技术领域technical field
本发明涉及图像处理领域，尤其涉及骑行者重识别技术。The invention relates to the field of image processing, in particular to the rider re-identification technology.
背景技术Background technique
随着城镇化率的升高，全国各城镇的常驻人口在逐渐增加，这给原本拥挤的交通带来了很大压力。各类交通违规行为也给相关治理部门带来麻烦。其中非机动车占机动车道的违规行为有多发、常发的特点。目前机动车的违规识别抓拍取证已基本做到高准确度、自动高效，而非机动车的治理目前比较低效。主要原因在于人车相关的检测、识别算法的研究非常欠缺。With the increase of urbanization rate, the resident population of cities and towns across the country is gradually increasing, which puts a lot of pressure on the originally congested traffic. All kinds of traffic violations have also brought troubles to relevant governance departments. Among them, violations of non-motor vehicles occupying motor vehicle lanes are multiple and frequent. At present, the illegal identification, capture and evidence collection of motor vehicles has basically achieved high accuracy, automatic and efficient, while the governance of non-motor vehicles is currently relatively inefficient. The main reason is that the research on human-vehicle-related detection and recognition algorithms is very lacking.
在实际治理中，机动车道上方会设点安装监控。通过对监控视频进行隔帧人车(骑行者与非机动车的组合体)检测，所有的非机动车占机动车道的违规行为将会被抓拍记录。但实际情况下只做到这步还不够，治理人员需要获取违规人车的前面与背面两张抓拍图片。人车的前面有骑行者的脸部信息，后面有电动车的牌照信息。这些信息能更好地辅助后续的违规处置。为达到这个要求，相关部门会将监控进行成对安装。一个摄像头朝车流方向拍摄(可抓拍背面)，另一个则朝相反方向拍摄(可抓拍正面)。无论是正面还是背面，人车检测都没有问题。难以处理的问题是，把一个监控下抓拍的正面人车与另一个监控下抓拍背面人车进行匹配，即骑行者重识别。实际场景下，载人、挡风物件等都会使得人车的前后两面相差巨大，这极大的增加了匹配困难。而且，可用于训练识别模型的数据极少。In actual governance, monitoring will be installed above the motorway. By detecting people and vehicles (a combination of cyclists and non-motor vehicles) at intervals in the surveillance video, all violations of non-motor vehicles occupying motor vehicle lanes will be captured and recorded. But in reality, it is not enough to just do this step. The management personnel need to obtain two snapshots of the front and back of the violating person's car. The front of the vehicle has the facial information of the rider, and the rear has the license plate information of the electric vehicle. This information can better assist subsequent violation handling. In order to meet this requirement, the relevant departments will install the monitors in pairs. One camera shoots in the direction of traffic (captures the rear) and the other in the opposite direction (captures the front). Whether it is the front or the back, there is no problem with the detection of people and vehicles. The intractable problem is to match the frontal person-vehicle captured under surveillance with another person-vehicle captured under surveillance, that is, rider re-identification. In actual scenarios, people and windshield objects will cause a huge difference between the front and rear sides of the vehicle, which greatly increases the difficulty of matching. Also, there is very little data available to train recognition models.
因此，需要提高骑行者重识别的匹配度。Therefore, it is necessary to improve the matching degree of rider re-identification.
发明内容Contents of the invention
本发明提供一种骑行者重识别方法、装置及设备，以提高骑行者重识别的匹配度。The invention provides a rider re-identification method, device and equipment to improve the matching degree of rider re-identification.
第一方面，提供了一种骑行者重识别方法，包括：In the first aspect, a rider re-identification method is provided, including:
对行人重识别模型进行迁移学习，得到骑行者重识别模型；Perform transfer learning on the pedestrian re-identification model to obtain the cyclist re-identification model;
采用所述骑行者重识别模型提取人车图片库中多个人车图片的特征向量；Using the re-identification model of the rider to extract the feature vectors of a plurality of pictures of people and cars in the picture library of people and cars;
计算所述特征向量之间的多个相似度值；calculating a plurality of similarity values between the feature vectors;
将最高的所述相似度值对应的至少两个所述人车图片识别为同一骑行者。Identifying at least two of the pictures of the person and the vehicle corresponding to the highest similarity value as the same cyclist.
在一个实现中，所述对行人重识别模型进行迁移学习，得到骑行者重识别模型，包括：In one implementation, the transfer learning is performed on the pedestrian re-identification model to obtain the cyclist re-identification model, including:
获取行人重识别模型的主干网络参数；Obtain the backbone network parameters of the pedestrian re-identification model;
根据骑行者训练样本的类别数，在所述行人重识别模型的主干网络后添加分类层；According to the number of categories of the rider training samples, a classification layer is added after the backbone network of the pedestrian re-identification model;
调整所述分类层的参数和所述主干网络参数，得到所述骑行者重识别模型。Adjusting the parameters of the classification layer and the parameters of the backbone network to obtain the rider re-identification model.
在又一个实现中，所述人车图片包括每一骑行者的正面人车图和背面人车图，所述采用所述骑行者重识别模型提取人车图片库中多个人车图片的特征向量，包括：In yet another implementation, the pictures of people and vehicles include the front and back pictures of people and vehicles of each rider, and the feature vectors of multiple pictures of people and vehicles in the library of pictures of people and vehicles are extracted by using the rider re-identification model. ,include:
分别提取所述正面人车图的特征向量和所述背面人车图的特征向量；Extracting the feature vectors of the frontal figure of people and vehicles and the feature vectors of the figure of people and vehicles on the back side respectively;
所述计算所述特征向量之间的多个相似度值，包括：The calculation of multiple similarity values between the feature vectors includes:
将所述正面人车图片的特征向量和所述背面人车图片的特征向量分别进行内积运算，得到所述多个特征向量之间的内积结果，其中，所述内积结果作为每一组正面人车图片的特征向量和背面人车图片的特征向量之间的相似度值。Perform inner product operations on the feature vectors of the front picture of people and vehicles and the feature vectors of the back picture of people and cars to obtain inner product results between the plurality of feature vectors, wherein the inner product results are used as each The similarity value between the eigenvectors of the picture of people and vehicles on the front and the eigenvectors of the pictures of people and vehicles on the back of the group.
在又一个实现中，所述方法还包括：In yet another implementation, the method also includes:
通过监控设备获取多个人车图片的图像信息，其中，所述监控设备至少包括一个对人车进行前方拍摄的摄像机，以及一个对人车进行后方拍摄的摄像机，所述图像信息包括所述人车图片的抓拍地点和抓拍时间；The image information of a plurality of pictures of people and vehicles is obtained through monitoring equipment, wherein the monitoring equipment includes at least one camera for shooting people and vehicles in front and one camera for taking pictures of people and vehicles behind, and the image information includes the people and vehicles The location and time of the capture of the picture;
根据所述抓拍地点和/或所述抓拍时间，将所述人车图片存储至相应类型的人车图片库；According to the snapping location and/or the snapping time, storing the pictures of people and vehicles in a library of pictures of people and cars of a corresponding type;
所述采用所述骑行者重识别模型提取人车图片库中中多个人车图片的特征向量，包括：The described use of the rider re-identification model to extract the feature vectors of multiple pictures of people and cars in the picture library of people and cars includes:
获取目标人车图片库，其中，所述目标人车图片库为所述用户指定类型的人车图片库；Obtaining a target person-vehicle image library, wherein the target person-vehicle image library is a user-specified type of person-vehicle image library;
采用所述骑行者重识别模型提取所述目标人车图片库中多个所述人车图片的特征向量。Using the rider re-identification model to extract feature vectors of multiple pictures of people and vehicles in the target library of pictures of people and vehicles.
在又一个实现中，所述方法还包括：In yet another implementation, the method also includes:
根据所述进行前方拍摄的摄像机以及所述进行后方拍摄的摄像机之间的位置距离，设置所述人车图片库中人车图片的存储时间；According to the position distance between the camera for shooting in front and the camera for shooting in the rear, the storage time of the pictures of people and cars in the picture library of people and cars is set;
当人车图片的存储时间超过所述设置的人车图片库中人车图片的存储时间时，将存储时间超过所述设置的存储时间的人车图片出库。When the storage time of the pictures of people and vehicles exceeds the storage time of the pictures of people and vehicles in the set picture library of people and vehicles, the pictures of people and vehicles whose storage time exceeds the set storage time are taken out of the warehouse.
第二方面，提供了一种骑行者重识别装置，包括：In the second aspect, a rider re-identification device is provided, including:
学习模块，用于对行人重识别模型进行迁移学习，得到骑行者重识别模型；The learning module is used to carry out transfer learning on the pedestrian re-identification model to obtain the cyclist re-identification model;
提取模块，用于采用所述骑行者重识别模型提取人车图片库中多个人车图片的特征向量；The extraction module is used to extract the feature vectors of multiple pictures of people and cars in the picture library of people and cars by using the re-identification model of the rider;
计算模块，用于计算所述特征向量之间的多个相似度值；A calculation module, configured to calculate a plurality of similarity values between the feature vectors;
识别模块，用于将最高的所述相似度值对应的至少两个所述人车图片识别为同一骑行者。An identification module, configured to identify at least two pictures of people and vehicles corresponding to the highest similarity value as the same cyclist.
在一个实现中，所述人车图片包括每一骑行者的正面人车图和背面人车图；In one implementation, the person-vehicle picture includes a front-side person-vehicle picture and a back-side person-vehicle picture of each cyclist;
所述提取模块具体用于分别提取所述正面人车图的特征向量和所述背面人车图的特征向量；The extraction module is specifically used to respectively extract the feature vectors of the frontal figure of people and vehicles and the feature vectors of the rear figure of people and vehicles;
所述计算模块具体用于将所述正面人车图片的特征向量和所述背面人车图片的特征向量分别进行内积运算，得到所述多个特征向量之间的内积结果，其中，所述内积结果作为每一组正面人车图片的特征向量和背面人车图片的特征向量之间的相似度值。The calculation module is specifically used to perform an inner product operation on the feature vector of the front picture of people and vehicles and the feature vector of the back picture of people and cars to obtain an inner product result between the multiple feature vectors, wherein the The above inner product result is used as the similarity value between the feature vectors of each group of frontal pictures of people and vehicles and the feature vectors of the back pictures of people and cars.
在又一个实现中，所述装置还包括：In yet another implementation, the device also includes:
获取模块，用于通过监控设备获取多个人车图片的图像信息，其中，所述监控设备至少包括一个对人车进行前方拍摄的摄像机，以及一个对人车进行后方拍摄的摄像机，所述图像信息包括所述人车图片的抓拍地点和抓拍时间；An acquisition module, configured to acquire image information of a plurality of pictures of people and vehicles through monitoring equipment, wherein the monitoring equipment at least includes a camera for shooting people and vehicles in front and a camera for taking pictures of people and vehicles in the rear, and the image information Including the capture location and capture time of the pictures of people and vehicles;
存储模块，用于根据所述抓拍地点和/或所述抓拍时间，将所述人车图片存储至相应类型的人车图片库；A storage module, configured to store the pictures of people and vehicles in a library of pictures of people and vehicles of a corresponding type according to the snapping location and/or the snapping time;
所述提取模块包括：The extraction module includes:
第一获取单元，用于获取目标人车图片库，其中，所述目标人车图片库为所述用户指定类型的人车图片库；The first acquiring unit is configured to acquire a target person-vehicle image library, wherein the target person-vehicle image library is a user-specified type of person-vehicle image library;
提取单元，用于采用所述骑行者重识别模型提取所述目标人车图片库中多个所述人车图片的特征向量。An extraction unit, configured to extract feature vectors of multiple pictures of people and vehicles in the target library of pictures of people and vehicles by using the rider re-identification model.
在又一个实现中，所述学习模块包括：In yet another implementation, the learning modules include:
第二获取单元，用于获取行人重识别模型的主干网络参数；The second acquisition unit is used to acquire the backbone network parameters of the pedestrian re-identification model;
添加单元，用于根据骑行者训练样本的类别数，在所述行人重识别模型的主干网络后添加分类层；Adding a unit for adding a classification layer after the backbone network of the pedestrian re-identification model according to the number of categories of the cyclist training samples;
调整单元，用于调整所述分类层的参数和所述主干网络参数，得到所述骑行者重识别模型。An adjustment unit, configured to adjust the parameters of the classification layer and the parameters of the backbone network to obtain the rider re-identification model.
在又一个实现中，所述装置还包括：In yet another implementation, the device also includes:
设置模块，用于根据所述进行前方拍摄的摄像机以及所述进行后方拍摄的摄像机之间的位置距离，设置所述人车图片库中人车图片的存储时间；The setting module is used to set the storage time of the pictures of people and vehicles in the picture library of people and vehicles according to the position distance between the camera for shooting in front and the camera for shooting in the rear;
所述存储模块还用于当人车图片的存储时间超过所述设置的人车图片库中人车图片的存储时间时，将存储时间超过所述设置的存储时间的人车图片出库。The storage module is further configured to, when the storage time of the pictures of people and vehicles exceeds the storage time of the pictures of people and vehicles in the set library of pictures of people and vehicles, take out the pictures of people and vehicles whose storage time exceeds the set storage time.
第三方面，提供了一种骑行者重识别设备，包括处理器、输入设备、输出设备和存储器，所述存储器用于存储计算机程序，所述计算机程序包括程序指令，所述处理器被配置用于调用所述程序指令，执行上述第一方面或第一方面中的任一种实现所述的方法。In a third aspect, a cyclist re-identification device is provided, including a processor, an input device, an output device, and a memory, the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to After invoking the program instructions, execute the above first aspect or any one of the first aspects to realize the described method.
第四方面，提供了一种计算机可读存储介质，所述计算机可读存储介质中存储有指令，当其在计算机上运行时，使得计算机执行上述第一方面或第一方面中的任一种实现所述的方法。In a fourth aspect, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, and when the computer-readable storage medium is run on a computer, the computer executes any one of the above-mentioned first aspect or the first aspect. implement the method described.
第五方面，提供了一种包含指令的计算机程序产品，当其在计算机上运行时，使得计算机执行上述第一方面或第一方面中的任一种实现所述的方法。In a fifth aspect, a computer program product containing instructions is provided, and when it is run on a computer, the computer executes the first aspect or any one of the first aspects to implement the method.
本发明实施例具有以下有益效果：Embodiments of the present invention have the following beneficial effects:
由于行人重识别在算法以及数据方面都有较好的积累，而骑行者重识别则是一个全新的问题，用于训练的样本有限，迁移学习能对行人重识别所学习的规律进行借鉴利用，因此通过对行人重识别模型进行迁移学习，得到骑行者重识别模型，利用该骑行者重识别模型提取人车图片库中多个人车图片的特征向量，计算特征向量之间的相似度值，并将最高的所述相似度值对应的至少两个所述人车图片识别为同一骑行者，提高了骑行者重识别的匹配度，实现对非机动车辆违法行为的监控。Since pedestrian re-identification has a good accumulation of algorithms and data, cyclist re-identification is a brand new problem, and the samples used for training are limited. Transfer learning can learn from the rules learned by pedestrian re-identification. Therefore, by performing transfer learning on the pedestrian re-identification model, a cyclist re-identification model is obtained, and the cyclist re-identification model is used to extract the feature vectors of multiple pictures of people and cars in the picture library of people and cars, and the similarity value between the feature vectors is calculated, and Identifying at least two pictures of people and vehicles corresponding to the highest similarity value as the same cyclist improves the matching degree of re-identification of cyclists and realizes the monitoring of illegal behavior of non-motor vehicles.
附图说明Description of drawings
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. Those skilled in the art can also obtain other drawings based on these drawings without creative work.
图1是本发明实施例示例的一种人车图片的采集示意图；Fig. 1 is a schematic diagram of collecting a picture of a person and a vehicle according to an example of an embodiment of the present invention;
图2是本发明实施例提供的一种骑行者重识别方法的流程示意图；Fig. 2 is a schematic flowchart of a rider re-identification method provided by an embodiment of the present invention;
图3是本发明实施例提供的又一种骑行者重识别方法的流程示意图；Fig. 3 is a schematic flowchart of another rider re-identification method provided by an embodiment of the present invention;
图4是本发明实施例提供的对行人重识别模型进行迁移学习的过程示意图；Fig. 4 is a schematic diagram of the process of performing transfer learning on the pedestrian re-identification model provided by the embodiment of the present invention;
图5是本发明实施例提供的一种骑行者重识别装置的结构示意图；Fig. 5 is a schematic structural diagram of a rider re-identification device provided by an embodiment of the present invention;
图6是本发明实施例提供的一种骑行者重识别设备的结构示意图。Fig. 6 is a schematic structural diagram of a rider re-identification device provided by an embodiment of the present invention.
具体实施方式Detailed ways
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.
为了获得人车更多的信息，一般地，交通监管部门会成对安装摄像机。如图1所示，是本发明实施例示例的一种人车图片的采集示意图，在机动车道上方的一定距离内安装监控设备对，包括一个对人车进行前方拍摄的摄像机1，以及一个对人车进行后方拍摄的摄像机2。摄像机1朝向车流的相反方向拍摄，可抓拍违法在机动车道行驶的非机动车和人的正面，摄像机2朝向车流方向拍摄，可抓拍非机动车和人的背面。然而，把一个监控下抓拍的正面人车与另一个监控下抓拍背面人车进行匹配，即骑行者重识别，目前没有相应的技术。实际场景下，载人、挡风物件等都会使得人车的前后两面相差巨大，这极大的增加了匹配困难。而且，可用于训练识别模型的数据极少。In order to obtain more information about people and vehicles, generally, traffic supervision departments will install cameras in pairs. As shown in Figure 1, it is a schematic diagram of a collection of pictures of people and vehicles according to an example of the embodiment of the present invention. A pair of monitoring equipment is installed within a certain distance above the motorway, including a camera 1 for taking pictures of people and vehicles ahead, and a pair of monitoring equipment. Camera 2 for people and vehicles to take pictures from behind. Camera 1 shoots in the opposite direction of the traffic flow, and can capture the front of non-motor vehicles and people illegally driving on the motorway. Camera 2 shoots in the direction of traffic flow, and can capture the back of non-motor vehicles and people. However, there is currently no corresponding technology to match a frontal person-vehicle captured under surveillance with another person-vehicle captured under surveillance, that is, rider re-identification. In actual scenarios, people and windshield objects will cause a huge difference between the front and rear sides of the vehicle, which greatly increases the difficulty of matching. Also, there is very little data available to train recognition models.
本发明实施例提供一种骑行者重识别方法、装置及设备，通过对行人重识别模型进行迁移学习，得到骑行者重识别模型，利用该骑行者重识别模型提取人车图片库中多个人车图片的特征向量，计算特征向量之间的相似度值，并将最高的所述相似度值对应的至少两个所述人车图片识别为同一骑行者，提高了骑行者重识别的匹配度，实现对非机动车辆违法行为的监控。Embodiments of the present invention provide a method, device, and equipment for re-identification of cyclists. By performing transfer learning on the re-identification model of pedestrians, a re-identification model of cyclists is obtained, and the re-identification model of cyclists is used to extract multiple people and vehicles in the image database of people and vehicles. The feature vector of the picture, calculating the similarity value between the feature vectors, and identifying at least two of the pictures of people and vehicles corresponding to the highest similarity value as the same rider, which improves the matching degree of rider re-identification, Realize the monitoring of illegal behavior of non-motor vehicles.
图2为本发明实施例提供的一种骑行者重识别方法的流程示意图，示例性的，该方法可包括以下步骤：Fig. 2 is a schematic flowchart of a rider re-identification method provided by an embodiment of the present invention. Exemplarily, the method may include the following steps:
S101、对行人重识别模型进行迁移学习，得到骑行者重识别模型。S101. Perform transfer learning on the pedestrian re-identification model to obtain a cyclist re-identification model.
行人重识别(Person re-identification)也称行人再识别，是利用计算机视觉技术判断图像或者视频序列中是否存在特定行人的技术。目前行人重识别的研究进展迅速，在几个公开数据集上，行人重识别的精度提高显著。行人重识别模型如残差网络ResNet-50。Person re-identification (Person re-identification), also known as pedestrian re-identification, is a technology that uses computer vision technology to determine whether a specific pedestrian exists in an image or video sequence. At present, the research on pedestrian re-identification is progressing rapidly. On several public datasets, the accuracy of pedestrian re-identification has improved significantly. Pedestrian re-identification models such as residual network ResNet-50.
由于行人重识别在算法以及数据方面都有较好的积累，而骑行者重识别则是一个全新的问题，用于训练的样本较少，迁移学习能对行人重识别所学习的规律进行借鉴利用。迁移学习是一种机器学习方法，由于数据集与现实数据之间的差异，导致在数据集A上训练好的模型在现实数据B上性能表现不佳。目前主要采用迁移学习的方法，在有标签的数据集A和无标签数据集B上训练，最后在数据集B的测试集上测试。也就是把为任务A开发的模型作为初始点，重新使用在为任务B开发模型的过程中。因此将行人重识别模型通过迁移学习改变为骑行者重识别模型是非常不错的选择，这样能解决骑行者重识别中的数据瓶颈问题。Since pedestrian re-identification has a good accumulation of algorithms and data, cyclist re-identification is a brand new problem, and there are few samples for training. Transfer learning can learn from the rules of pedestrian re-identification. . Transfer learning is a machine learning method where a model trained on dataset A performs poorly on real-world data B due to the discrepancy between the dataset and real-world data. At present, the transfer learning method is mainly used to train on the labeled data set A and unlabeled data set B, and finally test on the test set of data set B. That is, the model developed for task A is used as a starting point and reused in the process of developing the model for task B. Therefore, it is a very good choice to change the pedestrian re-identification model to the cyclist re-identification model through transfer learning, which can solve the data bottleneck problem in the cyclist re-identification.
S102、采用所述骑行者重识别模型提取人车图片库中多个人车图片的特征向量。S102. Using the rider re-identification model to extract feature vectors of multiple pictures of people and vehicles in the picture library of people and vehicles.
如图1所示，为了对违法骑行者进行监管，交通监管部门成对安装了摄像机，将这些摄像机摄取的人车图片存储到人车图片库中。具体地，该人车图片是从摄像机摄取的原始图片中检测并剪裁出来的。人车图片库中存储了多个人车图片。As shown in Figure 1, in order to supervise illegal cyclists, the traffic supervision department installed cameras in pairs, and stored the pictures of people and vehicles captured by these cameras in the picture library of people and vehicles. Specifically, the picture of the person and the vehicle is detected and cropped from the original picture captured by the camera. A number of pictures of people and cars are stored in the picture library of people and cars.
在获得上述骑行者重识别模型后，可以采用该骑行者重识别模型提取人车图片库中多个人车图片的特征向量，即将人车图片输入骑行者重识别模型，通过模型的复杂运算得到人车图片的特征向量。具体实现中，可以在人车图片库中获得多张人车图片，可以采用上述骑行者重识别模型逐一或者同时提取多张人车图片的特征向量。该人车图片的特征向量包括人的外貌、衣着、车牌、车的外在特征等。After obtaining the above cyclist re-identification model, the cyclist re-identification model can be used to extract the feature vectors of multiple pictures of people and cars in the picture library of people and cars, that is, the pictures of people and cars are input into the cyclist re-identification model, and the people and cars are obtained through the complex operation of the model. The feature vector of the car image. In a specific implementation, multiple pictures of people and vehicles can be obtained in the library of pictures of people and vehicles, and the feature vectors of multiple pictures of people and vehicles can be extracted one by one or simultaneously by using the above-mentioned rider re-identification model. The feature vector of the picture of the person and the car includes the person's appearance, clothing, license plate, external features of the car, and the like.
可选地，一个人车图片对应一条特征向量或一个类型的特征向量。例如，该特征向量为骑行者和车辆的外在特征。Optionally, a picture of a person and a vehicle corresponds to a feature vector or a type of feature vector. For example, the feature vectors are extrinsic features of the cyclist and the vehicle.
可选的，一个人车图片也对应多条特征向量或多个类型的特征向量。Optionally, a picture of a person and a vehicle also corresponds to multiple feature vectors or multiple types of feature vectors.
S103、计算所述特征向量之间的多个相似度值。S103. Calculate multiple similarity values between the feature vectors.
在提取出多个人车图片的特征向量后，计算特征向量之间的多个相似度值。具体地，分别计算任意两个或者多个特征向量的相似度值，得到提取出的所有特征向量之间的多个相似度值。可以根据需要设置进行匹配的特征向量个数。After extracting the feature vectors of multiple pictures of people and vehicles, multiple similarity values between the feature vectors are calculated. Specifically, the similarity values of any two or more feature vectors are respectively calculated to obtain multiple similarity values between all the extracted feature vectors. The number of eigenvectors for matching can be set as required.
例如，人车图片库中分别存储了骑行者的正面人车图片和背面人车图片，并且分别进行了标识，可以分别提取上述正面人车图片和背面人车图片的特征向量，并且计算任一正面人车图片的特征向量与提取的人车图片库中的所有的背面人车图片的特征向量之间的相似度值。For example, the person-vehicle image library stores the front image of the person-vehicle and the image of the rear person-vehicle of the cyclist respectively, and identifies them respectively. The feature vectors of the above-mentioned front image of the person-vehicle and the rear image of the person-vehicle can be extracted respectively, and any The similarity value between the eigenvectors of the frontal pictures of people and vehicles and the feature vectors of all the extracted pictures of people and vehicles on the back side.
可选地，上述步骤中提取了多个人车图片的一条特征向量或一个类型的特征向量，则计算这多个人车图片的特征向量之间的相似度，得到多个相似度值。Optionally, if a feature vector or a type of feature vector of multiple pictures of people and vehicles is extracted in the above steps, the similarity between the feature vectors of the pictures of people and vehicles is calculated to obtain multiple similarity values.
可选地，上述步骤中提取了多个类型的特征向量，则按照特征向量的类型，分别计算这多个人车图片的每个类型的特征向量之间的相似度。然后，可以按照每个类型的特征向量的权重，计算提取出的各个类型的特征向量之间的综合的相似度值，作为最终的相似度值。例如，对于人车图片A，提取出特征向量A1、A2；对于人车图片B，提取出特征向量B1、B2。而特征向量A1、B1为同一类型的特征向量，例如为骑行者的脸部特征向量；特征向量A2、B2为另一类型的特征向量，例如为骑行者的服装特征向量。设置特征向量A1、B1这一类特征向量的权重为0.7，设置特征向量A2、B2这一类特征向量的权重为0.3。则当计算出特征向量A1、B1之间的相似度值为98％，计算出特征向量A2、B2之间的相似度值为80％时，得到人车图片A、B的特征向量的综合的相似度为98％*0.7+80％*0.3。Optionally, if multiple types of feature vectors are extracted in the above steps, then according to the types of feature vectors, the similarity between each type of feature vectors of the multiple pictures of people and vehicles is calculated respectively. Then, according to the weight of each type of feature vector, a comprehensive similarity value between the extracted feature vectors of each type may be calculated as a final similarity value. For example, for the picture A of people and vehicles, the feature vectors A1 and A2 are extracted; for the picture B of people and cars, the feature vectors B1 and B2 are extracted. The feature vectors A1 and B1 are the same type of feature vectors, such as the rider's face feature vector; the feature vectors A2 and B2 are another type of feature vector, such as the rider's clothing feature vector. Set the weight of eigenvectors such as eigenvectors A1 and B1 to 0.7, and set the weight of eigenvectors such as eigenvectors A2 and B2 to 0.3. Then when the calculated similarity value between the feature vectors A1 and B1 is 98%, and the calculated similarity value between the feature vectors A2 and B2 is 80%, the comprehensive result of the feature vectors of the human-vehicle pictures A and B is obtained. The similarity is 98%*0.7+80%*0.3.
S104、将最高的所述相似度值对应的至少两个所述人车图片识别为同一骑行者。S104. Identify at least two pictures of people and vehicles corresponding to the highest similarity value as the same cyclist.
在计算出上述特征向量之间的多个相似度值之后，确定最高的相似度值，将最高的相似度值对应的至少两个人车图片识别为同一骑行者。这样，就可以获取该骑行者更多的特征信息，便于判断骑行者是否违法驾驶。After calculating a plurality of similarity values between the above feature vectors, determine the highest similarity value, and identify at least two pictures of people and vehicles corresponding to the highest similarity value as the same cyclist. In this way, more characteristic information of the cyclist can be obtained, which is convenient for judging whether the cyclist is driving illegally.
例如，人车图片库中分别存储了骑行者的正面人车图片和背面人车图片，并且分别进行了标识，在提取了任一正面人车图片的特征向量与提取的人车图片库中的所有的背面人车图片的特征向量之间的相似度值后，可确定与该正面人车图片之间相似度值最高的背面人车图片，从而将该组相似度值最高的正面人车图片和背面人车图片识别为同一骑行者。假设该骑行者违法驾驶，但根据抓拍到的正面人车图片只能获取该骑行者和车辆的正面，但获取不到车辆的车牌信息，通过本申请的方法，将匹配到与正面人车图片相似度最高的背面人车图片，可以确定该组正面人车图片和背面人车图片为同一骑行者，则可以根据该背面人车图片获取到车牌信息，从而获取到了该骑行者更多的特征信息，便于判断骑行者是否违法驾驶。For example, the person-vehicle picture library stores the rider's front-side person-vehicle picture and the back-side person-vehicle picture respectively, and identifies them respectively. After the similarity values between the eigenvectors of all the pictures of people and vehicles on the back side, the picture of people and vehicles on the back with the highest similarity value between the pictures of people and vehicles on the front can be determined, so that the group of pictures of people and cars on the front with the highest similarity value It is identified as the same rider as the picture of the person and vehicle on the back. Assuming that the cyclist is driving illegally, but only the front of the cyclist and the vehicle can be obtained according to the captured frontal picture of the person and the vehicle, but the license plate information of the vehicle cannot be obtained. Through the method of this application, it will be matched with the frontal picture of the person and the vehicle. For the picture of the person and the vehicle on the back with the highest similarity, it can be determined that the group of pictures of the person and the vehicle on the front and the picture of the person and the vehicle on the back are the same rider, and then the license plate information can be obtained based on the picture of the person and the vehicle on the back, so as to obtain more characteristics of the rider The information is convenient for judging whether the cyclist is driving illegally.
根据本发明实施例提供的一种骑行者重识别方法，通过对行人重识别模型进行迁移学习，得到骑行者重识别模型，利用该骑行者重识别模型提取人车图片库中多个人车图片的特征向量，计算特征向量之间的相似度值，并将最高的所述相似度值对应的至少两个所述人车图片识别为同一骑行者，提高了骑行者重识别的匹配度，实现对非机动车辆违法行为的监控。According to a cyclist re-identification method provided in an embodiment of the present invention, the cyclist re-identification model is obtained by performing transfer learning on the pedestrian re-identification model, and the cyclist re-identification model is used to extract the images of multiple people and vehicles in the person-vehicle image library. Eigenvectors, calculating the similarity value between the eigenvectors, and identifying at least two of the pictures of people and vehicles corresponding to the highest similarity value as the same cyclist, which improves the matching degree of re-identification of the cyclist, and realizes Monitoring of non-motor vehicle violations.
图3为本发明实施例提供的一种骑行者重识别方法的流程示意图，示例性的，该方法可包括以下步骤：Fig. 3 is a schematic flowchart of a rider re-identification method provided by an embodiment of the present invention. Exemplarily, the method may include the following steps:
S201、根据进行前方拍摄的摄像机以及进行后方拍摄的摄像机之间的位置距离，设置人车图片库中人车图片的存储时间。S201. According to the positional distance between the camera for shooting in front and the camera for shooting in the rear, set the storage time of the pictures of people and vehicles in the picture library of people and vehicles.
由于机动车道上设置的摄像机每天/每小时/每分钟都要拍摄和存储大量的照片，因此，为了节省存储空间，需要设置人车图片库中人车图片的存储时间，超出存储时间的人车图片删除出库。Since the cameras set up on the motorway take and store a large number of photos every day/hour/minute, in order to save storage space, it is necessary to set the storage time of the pictures of people and cars in the picture library of people and cars. The picture is deleted from the library.
一般地，进行前方拍摄的摄像机和进行后方拍摄的摄像机之间有一定的位置距离，因此，获得正面拍摄照片和反面拍摄照片之间有一定的时间差，然而，正面拍摄照片和反面拍摄照片是相似度最高的人车图片组，因此，需要将正面拍摄照片和反面拍摄照片同时保留在人车图片库，以便于后续的匹配。因此，可以根据进行前方拍摄的摄像机以及进行后方拍摄的摄像机之间的位置距离，设置人车图片库的存储时间，或者称人车图片库的持续存储时长(最新入库人车图片与库中最早的人车图片的时间差)。Generally, there is a certain positional distance between the camera that takes the front shot and the camera that takes the rear shot, so there is a certain time lag between getting the front shot and the back shot, however, the front shot and the back shot are similar Therefore, it is necessary to keep the front and back photos in the person-vehicle image library at the same time, so as to facilitate subsequent matching. Therefore, according to the positional distance between the camera that takes the front shot and the camera that takes the rear shot, the storage time of the person-vehicle picture library can be set, or the continuous storage time of the person-vehicle picture library (the latest picture of the person and car in the warehouse is the same as that in the library). The time difference of the earliest pictures of people and vehicles).
S202、通过监控设备获取多个人车图片的图像信息。S202. Obtain image information of a plurality of pictures of people and vehicles through a monitoring device.
如图1所示，交通监管部门在道路上方设置监控设备，可以通过监控设备摄取人车图片。其中，该监控设备至少包括一个对人车进行前方拍摄的摄像机，以及一个对人车进行后方拍摄的摄像机，从而可以获取骑行者的正面人车图片和背面人车图片。As shown in Figure 1, the traffic supervision department has set up monitoring equipment above the road, and can capture pictures of people and vehicles through the monitoring equipment. Wherein, the monitoring equipment includes at least one camera for shooting the front of the person and the vehicle, and one camera for shooting the rear of the person and the vehicle, so as to obtain the front and rear pictures of the person and the vehicle of the rider.
在存储人车图片的同时，还可以获取人车图片的图像信息。其中，该图像信息包括人车图片的抓拍地点和抓拍时间。While storing the pictures of people and vehicles, the image information of the pictures of people and vehicles can also be obtained. Wherein, the image information includes the location and time of capturing the pictures of people and vehicles.
S203、根据所述抓拍地点和/或所述抓拍时间，将所述人车图片存储至相应类型的人车图片库。S203. According to the snapshot location and/or the snapshot time, store the pictures of people and vehicles in a library of pictures of people and vehicles of a corresponding type.
本实施例中，为了提高匹配效率，缩小人车图片库的图片来源范围。具体地，根据人车图片源对应的监控设备抓拍地点、抓拍时间、或者抓拍地点和抓拍时间来划分人车图片库的类型。即一个人车图片库只存储某个抓拍地点的若干个监控设备的人车图片，或者存储某一个抓拍时间范围内获得的人车图片，或者存储某个抓拍地点的某一个抓拍时间范围内获得的人车图片。In this embodiment, in order to improve the matching efficiency, the source range of pictures in the human-vehicle picture library is narrowed down. Specifically, the type of the person-vehicle picture library is classified according to the monitoring device capture location and capture time corresponding to the source of the person-vehicle picture, or the capture location and capture time. That is, a person-vehicle picture library only stores the pictures of people and vehicles of several monitoring devices at a certain capture location, or stores the pictures of people and vehicles obtained within a certain capture time range, or stores the images obtained within a certain capture time range of a certain capture location pictures of people and cars.
例如，在某一车道上间隔设置了十个摄像机，其中，摄像机A和摄像机B是一个监控设备对，即摄像机A负责拍摄人车正面，摄像机B负责拍摄人车背面；摄像机C和摄像机D是一个监控设备对，以此类推。则可以将摄像机A和摄像机B拍摄的照片存储至人车图片库1，将摄像机C和摄像机D存储至人车图片库2，等等。又例如，摄像机A和摄像机B分别在8:00～9:00拍摄得到多个人车图片，又分别在9:00～10:00拍摄得到多个人车图片，则可以将摄像机A和摄像机B在8:00～9:00拍摄得到的多个人车图片存储至人车图片库1，将摄像机A和摄像机B在9:00～10:00拍摄得到的多个人车图片存储至人车图片库2，以此类推。For example, ten cameras are set at intervals on a certain lane, among which, camera A and camera B are a monitoring device pair, that is, camera A is responsible for photographing the front of the person and the vehicle, and camera B is responsible for photographing the back of the person and the vehicle; camera C and camera D are A monitoring device pair, and so on. Then, the photos taken by camera A and camera B can be stored in the person-vehicle image library 1, and the photos taken by camera C and camera D can be stored in the person-vehicle image library 2, and so on. For another example, camera A and camera B take multiple pictures of people and vehicles at 8:00-9:00 respectively, and then take multiple pictures of people and cars at 9:00-10:00 respectively, then camera A and camera B can be placed at The pictures of multiple people and vehicles taken from 8:00 to 9:00 are stored in the picture library 1 of people and vehicles, and the pictures of multiple people and cars taken by camera A and camera B between 9:00 and 10:00 are stored in the picture library of people and vehicles 2 , and so on.
S204、获取行人重识别模型的主干网络参数。S204. Obtain backbone network parameters of the pedestrian re-identification model.
S205、根据骑行者训练样本的类别数，在所述行人重识别模型的主干网络后添加分类层。S205. Add a classification layer after the backbone network of the pedestrian re-identification model according to the number of categories of the cyclist training samples.
S206、调整所述分类层的参数和所述主干网络参数，得到所述骑行者重识别模型。S206. Adjust the parameters of the classification layer and the parameters of the backbone network to obtain the rider re-identification model.
上述步骤S204～S206为对行人重识别模型进行迁移学习，得到骑行者重识别模型。其原理为将行人重识别模型视为骑行者重识别模型的初始模型，通过若干骑行者训练样本的训练，得到较为精确的骑行者重识别模型。如图4所示的本发明实施例提供的对行人重识别模型进行迁移学习的过程示意图，具体的训练过程为：首先，获取一个训练过的行人重识别模型。然后，获取该模型的主干网络参数(Resnet-backbone)(一种残差网络)，并根据骑行者训练样本的类别数，在该行人重识别模型的主干网络后添加分类层(FC)。例如，人工将10万个骑行者训练样本进行归类，属于同一个骑行者的训练样本归为一类，将上述10万个骑行者训练样本归为27000类，则根据训练样本的类别数，添加分类层，这样训练后的分类层的输出向量为27000维的向量。最后，根据损失函数(softmaxloss)调整参数，先调整分类层的参数，待损失稳定后再调整整个网络参数，包括分类层的参数和主干网络参数，得到骑行者重识别模型。The above steps S204-S206 are performing transfer learning on the pedestrian re-identification model to obtain the cyclist re-identification model. The principle is to regard the pedestrian re-identification model as the initial model of the cyclist re-identification model, and obtain a more accurate cyclist re-identification model through the training of several cyclist training samples. As shown in FIG. 4 , a schematic diagram of the process of performing transfer learning on the pedestrian re-identification model provided by the embodiment of the present invention, the specific training process is as follows: first, a trained pedestrian re-identification model is acquired. Then, obtain the backbone network parameters (Resnet-backbone) of the model (a residual network), and add a classification layer (FC) after the backbone network of the pedestrian re-identification model according to the number of categories of the cyclist training samples. For example, manually classify 100,000 cyclist training samples, and classify the training samples belonging to the same cyclist into one category, and classify the above 100,000 cyclist training samples into 27,000 categories, then according to the number of categories of training samples, Add a classification layer so that the output vector of the trained classification layer is a 27000-dimensional vector. Finally, adjust the parameters according to the loss function (softmaxloss), first adjust the parameters of the classification layer, and then adjust the parameters of the entire network after the loss is stable, including the parameters of the classification layer and the backbone network parameters, and obtain the rider re-identification model.
S207、获取目标人车图片库。S207. Obtain a target person-vehicle image library.
根据上述人车图片库的存储分类，用户指定对某个类型的人车图片库进行特征向量的提取和匹配，即指定对某个抓拍地点和/或某个抓拍时间的人车图片进行特征向量的提取和匹配。因此，获取目标人车图片库，该目标人车图片库为所述用户指定类型的人车图片库。According to the storage classification of the above-mentioned person-vehicle image library, the user specifies to extract and match feature vectors for a certain type of person-vehicle image library, that is, to specify the feature vector for a certain capture location and/or a certain capture time. extraction and matching. Therefore, a target person-vehicle image library is acquired, and the target person-vehicle image library is a type of person-vehicle image library specified by the user.
S208、采用所述骑行者重识别模型提取所述目标人车图片库中多个所述人车图片的特征向量。S208. Using the rider re-identification model to extract feature vectors of multiple pictures of people and cars in the target library of pictures of people and cars.
在获取目标人车图片库后，采用骑行者重识别模型提取该目标人车图片库中多个人车图片的特征向量。After obtaining the target person-vehicle image library, the cyclist re-identification model is used to extract the feature vectors of multiple person-vehicle images in the target person-vehicle image library.
具体地，如下面的公式1所示：Specifically, as shown in Equation 1 below:
feati＝net(xi)……公式1feati ＝net(xi )...Formula 1
其中，xi为第i张人车图片，feati为对应的人车特征向量。net即为上述残差网络的函数。Among them, xi is the i-th picture of people and vehicles, and feati is the corresponding feature vector of people and vehicles. net is a function of the above residual network.
在具体的示例中，人车图片可以包括每一骑行者的正面人车图和背面人车图，则S208包括：分别提取所述正面人车图的特征向量和所述背面人车图的特征向量。In a specific example, the person-vehicle picture may include the front-side person-vehicle image and the back-side image of the person-vehicle of each cyclist, then S208 includes: extracting the feature vectors of the front person-vehicle image and the features of the rear person-vehicle image respectively vector.
S209、计算所述特征向量之间的多个相似度。S209. Calculate multiple similarities between the feature vectors.
在提取出多个人车图片的特征向量后，计算特征向量之间的多个相似度值。具体地，分别计算任意两个或者多个特征向量的相似度值，得到提取出的所有特征向量之间的多个相似度值。可以根据需要设置进行匹配的特征向量个数。After extracting the feature vectors of multiple pictures of people and vehicles, multiple similarity values between the feature vectors are calculated. Specifically, the similarity values of any two or more feature vectors are respectively calculated to obtain multiple similarity values between all the extracted feature vectors. The number of eigenvectors for matching can be set as required.
在具体的示例中，人车图片可以包括每一骑行者的正面人车图和背面人车图，S209具体包括：将所述正面人车图片的特征向量和所述背面人车图片的特征向量分别进行内积运算，得到所述多个特征向量之间的内积结果，其中，所述内积结果作为每一组正面人车图片的特征向量和背面人车图片的特征向量之间的相似度值。具体地，可以采用下面的公式2得到上述相似度值：In a specific example, the pictures of people and vehicles may include the front and back pictures of people and cars of each rider, and S209 specifically includes: combining the feature vectors of the front pictures of people and cars with the feature vectors of the back pictures of people and cars Carry out the inner product operation respectively, obtain the inner product result between described multiple eigenvectors, wherein, described inner product result is as the similarity between the eigenvectors of each group of front people-vehicle picture and the feature vector of back side people-vehicle picture degree value. Specifically, the above similarity value can be obtained by using the following formula 2:
similarity＝InnerProduct(feati,featj)……公式2similarity＝InnerProduct(feati , featj )...Formula 2
其中，InnerProduct是指两个特征向量间的内积结果。Among them, InnerProduct refers to the inner product result between two feature vectors.
S210、将最高的所述相似度值对应的至少两个所述人车图片识别为同一骑行者。S210. Identify at least two of the pictures of people and vehicles corresponding to the highest similarity value as the same cyclist.
在计算出上述特征向量之间的多个相似度值之后，确定最高的相似度值，将最高的相似度值对应的至少两个人车图片识别为同一骑行者。这样，就可以获取该骑行者更多的特征信息，便于判断骑行者是否违法驾驶。After calculating a plurality of similarity values between the above feature vectors, determine the highest similarity value, and identify at least two pictures of people and vehicles corresponding to the highest similarity value as the same cyclist. In this way, more characteristic information of the cyclist can be obtained, which is convenient for judging whether the cyclist is driving illegally.
具体地，可以采用下面的公式3确定相似度最高的特征向量：Specifically, the following formula 3 can be used to determine the feature vector with the highest similarity:
其中，根据最大自变量点集函数argmax可以得到使得InnerProduct(feati,featj)取得最大值所对应的变量点feati和featj，从而确定相似度最高的至少两个特征向量。Among them, according to the function argmax of the largest independent variable point set, the variable points feati and featj corresponding to the maximum value of InnerProduct(feati , featj ) can be obtained, so as to determine at least two feature vectors with the highest similarity.
在确定了相似度最高的至少两个特征向量后，将至少两个特征向量对应的至少两个人车图片识别为同一骑行者。After at least two feature vectors with the highest similarity are determined, at least two pictures of people and vehicles corresponding to the at least two feature vectors are identified as the same rider.
S211、当人车图片的存储时间超过所述设置的人车图片库中人车图片的存储时间时，将存储时间超过所述设置的存储时间的人车图片出库。S211. When the storage time of the pictures of people and vehicles exceeds the storage time of the pictures of people and vehicles in the set library of pictures of people and vehicles, take out the pictures of people and vehicles whose storage time exceeds the set storage time.
为了节省存储空间，根据上述步骤S201中设置的存储时间或者称存储持续时长，在判断人车图片的存储时间超过上述设置的人车图片库中人车图片的存储时间时，将存储时间超过上述设置的存储时间的人车图片出库。In order to save storage space, according to the storage time or storage duration set in the above step S201, when it is judged that the storage time of the pictures of people and vehicles exceeds the storage time of the pictures of people and vehicles in the picture library of people and vehicles set above, the storage time will exceed the above-mentioned The pictures of people and vehicles in the set storage time are out of the warehouse.
根据本发明实施例提供的一种骑行者重识别方法，通过对行人重识别模型进行迁移学习，得到骑行者重识别模型，利用该骑行者重识别模型提取人车图片库中多个人车图片的特征向量，计算特征向量之间的相似度值，并将最高的所述相似度值对应的至少两个所述人车图片识别为同一骑行者，提高了骑行者重识别的匹配度，实现对非机动车辆违法行为的监控；且根据抓拍时间和/或抓拍地点，选取目标人车图片库，提取目标人车图片库中的人车图片的特征向量进行相似度匹配，以提高人车图片的特征向量匹配的效率；且对存储超时的人车图片及时出库，以节省存储空间。According to a cyclist re-identification method provided in an embodiment of the present invention, the cyclist re-identification model is obtained by performing transfer learning on the pedestrian re-identification model, and the cyclist re-identification model is used to extract the images of multiple people and vehicles in the person-vehicle image library. Eigenvectors, calculating the similarity value between the eigenvectors, and identifying at least two of the pictures of people and vehicles corresponding to the highest similarity value as the same cyclist, which improves the matching degree of re-identification of the cyclist, and realizes Monitoring of illegal behavior of non-motor vehicles; and according to the capture time and/or capture location, select the target person-vehicle image library, extract the feature vectors of the person-vehicle images in the target person-vehicle image library for similarity matching, in order to improve the accuracy of person-vehicle images. The efficiency of eigenvector matching; and timely release of pictures of people and vehicles that have been stored over time to save storage space.
基于上述实施例中的骑行者重识别方法的同一构思，如图5所示，本发明实施例还提供一种骑行者重识别装置100，该装置可应用于上述图2、图3所述的骑行者重识别方法中。该装置100包括：学习模块11、提取模块12、计算模块13和识别模块14，还可以包括获取模块15、存储模块16和设置模块17。示例性的：Based on the same idea of the rider re-identification method in the above-mentioned embodiments, as shown in FIG. 5, the embodiment of the present invention also provides a rider re-identification device 100, which can be applied to the above-mentioned FIG. 2 and FIG. 3 In the rider re-identification method. The device 100 includes: a learning module 11 , an extraction module 12 , a calculation module 13 and an identification module 14 , and may also include an acquisition module 15 , a storage module 16 and a setting module 17 . Exemplary:
学习模块11，用于对行人重识别模型进行迁移学习，得到骑行者重识别模型；The learning module 11 is used to carry out transfer learning to the pedestrian re-identification model to obtain the cyclist re-identification model;
提取模块12，用于采用所述骑行者重识别模型提取人车图片库中多个人车图片的特征向量；Extraction module 12, for adopting described cyclist re-identification model to extract the feature vector of a plurality of people-vehicle pictures in the people-vehicle picture library;
计算模块13，用于计算所述特征向量之间的多个相似度值；Calculation module 13, for calculating a plurality of similarity values between the feature vectors;
识别模块14，用于将最高的所述相似度值对应的至少两个所述人车图片识别为同一骑行者。The identification module 14 is configured to identify at least two pictures of people and vehicles corresponding to the highest similarity value as the same rider.
在一个实现中，所述人车图片包括每一骑行者的正面人车图和背面人车图；In one implementation, the person-vehicle picture includes a front-side person-vehicle picture and a back-side person-vehicle picture of each cyclist;
所述提取模块12具体用于分别提取所述正面人车图的特征向量和所述背面人车图的特征向量；The extraction module 12 is specifically used to extract the feature vectors of the front people-vehicle diagram and the feature vectors of the back people-vehicle diagram;
所述计算模块13具体用于将所述正面人车图片的特征向量和所述背面人车图片的特征向量分别进行内积运算，得到所述多个特征向量之间的内积结果，其中，所述内积结果作为每一组正面人车图片的特征向量和背面人车图片的特征向量之间的相似度值。The calculation module 13 is specifically used to perform an inner product operation on the feature vectors of the front picture of people and vehicles and the feature vector of the back picture of people and cars to obtain the inner product result between the multiple feature vectors, wherein, The inner product result is used as the similarity value between the feature vectors of each group of frontal pictures of people and vehicles and the feature vectors of the back pictures of people and vehicles.
在又一个实现中，所述装置还包括：In yet another implementation, the device also includes:
获取模块15，用于通过监控设备获取多个人车图片的图像信息，其中，所述监控设备对至少包括一个对人车进行前方拍摄的摄像机，以及一个对人车进行后方拍摄的摄像机，所述图像信息包括所述人车图片的抓拍地点和抓拍时间；The acquisition module 15 is used to acquire image information of a plurality of pictures of people and vehicles through monitoring equipment, wherein the monitoring equipment includes at least one camera for shooting people and vehicles in front and one camera for taking pictures of people and vehicles in the rear. The image information includes the capture location and capture time of the pictures of people and vehicles;
存储模块16，用于根据所述抓拍地点和/或所述抓拍时间，将所述人车图片存储至相应类型的人车图片库；A storage module 16, configured to store the pictures of people and vehicles in a library of pictures of people and vehicles of a corresponding type according to the snapping location and/or the snapping time;
所述提取模块12包括：Described extraction module 12 comprises:
第一获取单元121，用于获取目标人车图片库，其中，所述目标人车图片库为所述用户指定类型的人车图片库；The first acquiring unit 121 is configured to acquire a target person-vehicle image library, wherein the target person-vehicle image library is a user-specified type of person-vehicle image library;
提取单元122，用于采用所述骑行者重识别模型提取所述目标人车图片库中多个所述人车图片的特征向量。The extraction unit 122 is configured to use the rider re-identification model to extract feature vectors of multiple pictures of people and vehicles in the target picture library of people and vehicles.
在又一个实现中，所述学习模块11包括：In yet another implementation, the learning module 11 includes:
第二获取单元111，用于获取行人重识别模型的主干网络参数；The second obtaining unit 111 is used to obtain the backbone network parameters of the pedestrian re-identification model;
添加单元112，用于根据骑行者训练样本的类别数，在所述行人重识别模型的主干网络后添加分类层；Adding unit 112, for adding classification layer after the backbone network of described pedestrian re-identification model according to the category number of cyclist training samples;
调整单元113，用于调整所述分类层的参数和所述主干网络参数，得到所述骑行者重识别模型。The adjustment unit 113 is configured to adjust the parameters of the classification layer and the parameters of the backbone network to obtain the rider re-identification model.
在又一个实现中，所述装置还包括：In yet another implementation, the device also includes:
设置模块17，用于根据所述进行前方拍摄的摄像机以及所述进行后方拍摄的摄像机之间的位置距离，设置所述人车图片库中人车图片的存储时间；The setting module 17 is used to set the storage time of the pictures of people and vehicles in the picture library of people and vehicles according to the position distance between the camera for shooting in front and the camera for shooting in the rear;
所述存储模块16还用于当人车图片的存储时间超过所述设置的人车图片库中人车图片的存储时间时，将存储时间超过所述设置的存储时间的人车图片出库。The storage module 16 is also used for, when the storage time of the pictures of people and vehicles exceeds the storage time of the pictures of people and vehicles in the set library of pictures of people and vehicles, take out the pictures of people and vehicles whose storage time exceeds the set storage time.
有关上述各个模块、单元更详细的描述可以参考上述图2、图3所述的骑行者重识别方法的相关描述得到，这里不加赘述。A more detailed description of the above-mentioned modules and units can be obtained by referring to the related description of the rider re-identification method described above in FIG. 2 and FIG. 3 , and will not be repeated here.
根据本发明实施例提供的一种骑行者重识别装置，通过对行人重识别模型进行迁移学习，得到骑行者重识别模型，利用该骑行者重识别模型提取人车图片库中多个人车图片的特征向量，计算特征向量之间的相似度值，并将最高的所述相似度值对应的至少两个所述人车图片识别为同一骑行者，提高了骑行者重识别的匹配度，实现对非机动车辆违法行为的监控；且根据抓拍时间和/或抓拍地点，选取目标人车图片库，提取目标人车图片库中的人车图片的特征向量进行相似度匹配，以提高人车图片的特征向量匹配的效率；且对存储超时的人车图片及时出库，以节省存储空间。According to a cyclist re-identification device provided in an embodiment of the present invention, the cyclist re-identification model is obtained by performing transfer learning on the pedestrian re-identification model, and the cyclist re-identification model is used to extract the images of multiple people and vehicles in the person-vehicle image database Eigenvectors, calculating the similarity value between the eigenvectors, and identifying at least two of the pictures of people and vehicles corresponding to the highest similarity value as the same cyclist, which improves the matching degree of re-identification of the cyclist, and realizes Monitoring of illegal behavior of non-motor vehicles; and according to the capture time and/or capture location, select the target person-vehicle image library, extract the feature vectors of the person-vehicle images in the target person-vehicle image library for similarity matching, in order to improve the accuracy of person-vehicle images. The efficiency of eigenvector matching; and timely release of pictures of people and vehicles that have been stored over time to save storage space.
图6为本发明实施例提供的一种骑行者重识别设备的结构示意图，该设备200包括：包括处理器21，还可包括输入装置22、输出装置23和存储器24。该输入装置22、输出装置23、存储器24和处理器21之间通过总线相互连接。FIG. 6 is a schematic structural diagram of a rider re-identification device provided by an embodiment of the present invention. The device 200 includes: a processor 21 , and may also include an input device 22 , an output device 23 and a memory 24 . The input device 22 , the output device 23 , the memory 24 and the processor 21 are connected to each other through a bus.
存储器24包括但不限于是随机存储记忆体(random access memory，RAM)、只读存储器(read-only memory，ROM)、可擦除可编程只读存储器(erasable programmable readonly memory，EPROM)、或便携式只读存储器(compact disc read-only memory，CD-ROM)，该存储器用于相关指令及数据。Memory 24 includes, but is not limited to, random access memory (random access memory, RAM), read-only memory (read-only memory, ROM), erasable programmable read-only memory (erasable programmable readonly memory, EPROM), or portable Read-only memory (compact disc read-only memory, CD-ROM), which is used for related instructions and data.
输入装置22用于输入数据和/或信号，以及输出装置23用于输出数据和/或信号。输出装置和输入装置可以是独立的器件，也可以是一个整体的器件。The input device 22 is used for inputting data and/or signals, and the output device 23 is used for outputting data and/or signals. The output device and the input device can be independent devices or an integrated device.
处理器21可以包括是一个或多个处理器，例如包括一个或多个中央处理器(central processing unit，CPU)，在处理器是一个CPU的情况下，该CPU可以是单核CPU，也可以是多核CPU。The processor 21 may include one or more processors, such as one or more central processing units (central processing unit, CPU). In the case where the processor is a CPU, the CPU may be a single-core CPU, or It is a multi-core CPU.
存储器24用于存储网络设备的程序代码和数据。The memory 24 is used to store program codes and data of the network device.
处理器21用于调用该存储器中的程序代码和数据，执行如下步骤：The processor 21 is used to call the program code and data in the memory, and performs the following steps:
对行人重识别模型进行迁移学习，得到骑行者重识别模型；Perform transfer learning on the pedestrian re-identification model to obtain the cyclist re-identification model;
采用所述骑行者重识别模型提取人车图片库中多个人车图片的特征向量；Using the re-identification model of the rider to extract the feature vectors of a plurality of pictures of people and cars in the picture library of people and cars;
计算所述特征向量之间的多个相似度值；calculating a plurality of similarity values between the feature vectors;
将最高的所述相似度值对应的至少两个所述人车图片识别为同一骑行者。Identifying at least two of the pictures of the person and the vehicle corresponding to the highest similarity value as the same cyclist.
在一个实现中，所述处理器21执行所述对行人重识别模型进行迁移学习，得到骑行者重识别模型的步骤，包括：In one implementation, the processor 21 executes the step of performing transfer learning on the pedestrian re-identification model to obtain the cyclist re-identification model, including:
获取行人重识别模型的主干网络参数；Obtain the backbone network parameters of the pedestrian re-identification model;
根据骑行者训练样本的类别数，在所述行人重识别模型的主干网络后添加分类层；According to the number of categories of the rider training samples, a classification layer is added after the backbone network of the pedestrian re-identification model;
调整所述分类层的参数和所述主干网络参数，得到所述骑行者重识别模型。Adjusting the parameters of the classification layer and the parameters of the backbone network to obtain the rider re-identification model.
在又一个实现中，所述人车图片包括每一骑行者的正面人车图和背面人车图，所述处理器21执行所述采用所述骑行者重识别模型提取人车图片库中多个人车图片的特征向量的步骤，包括：In yet another implementation, the person-vehicle picture includes a front-side person-vehicle image and a back-side person-vehicle image of each cyclist, and the processor 21 executes the process of using the rider re-identification model to extract multiple images from the person-vehicle image library. The steps of the feature vector of the personal car picture include:
分别提取所述正面人车图的特征向量和所述背面人车图的特征向量；Extracting the feature vectors of the frontal figure of people and vehicles and the feature vectors of the figure of people and vehicles on the back side respectively;
所述处理器21执行所述计算所述特征向量之间的多个相似度值的步骤，包括：The processor 21 executes the step of calculating a plurality of similarity values between the feature vectors, including:
将所述正面人车图片的特征向量和所述背面人车图片的特征向量分别进行内积运算，得到所述多个特征向量之间的内积结果，其中，所述内积结果作为每一组正面人车图片的特征向量和背面人车图片的特征向量之间的相似度值。Perform inner product operations on the feature vectors of the front picture of people and vehicles and the feature vectors of the back picture of people and cars to obtain inner product results between the plurality of feature vectors, wherein the inner product results are used as each The similarity value between the eigenvectors of the picture of people and vehicles on the front and the eigenvectors of the pictures of people and vehicles on the back of the group.
在又一个实现中，所述处理器21还执行如下步骤：In yet another implementation, the processor 21 also performs the following steps:
通过监控设备获取多个人车图片的图像信息，其中，所述监控设备对至少包括一个对人车进行前方拍摄的摄像机，以及一个对人车进行后方拍摄的摄像机，所述图像信息包括所述人车图片的抓拍地点和抓拍时间；The image information of a plurality of pictures of people and vehicles is obtained by monitoring equipment, wherein the monitoring equipment includes at least one camera for shooting people and vehicles in front and one camera for taking pictures of people and vehicles in the rear, and the image information includes the people and vehicles The capture location and capture time of the car picture;
根据所述抓拍地点和/或所述抓拍时间，将所述人车图片存储至相应类型的人车图片库；According to the snapping location and/or the snapping time, storing the pictures of people and vehicles in a library of pictures of people and cars of a corresponding type;
所述处理器21执行所述采用所述骑行者重识别模型提取人车图片库中中多个人车图片的特征向量的步骤，包括：The processor 21 executes the step of using the rider re-identification model to extract the feature vectors of multiple pictures of people and cars in the picture library of people and cars, including:
获取目标人车图片库，其中，所述目标人车图片库为所述用户指定类型的人车图片库；Obtaining a target person-vehicle image library, wherein the target person-vehicle image library is a user-specified type of person-vehicle image library;
采用所述骑行者重识别模型提取所述目标人车图片库中多个所述人车图片的特征向量。Using the rider re-identification model to extract feature vectors of multiple pictures of people and vehicles in the target library of pictures of people and vehicles.
在又一个实现中，所述处理器21还执行如下步骤：In yet another implementation, the processor 21 also performs the following steps:
根据所述进行前方拍摄的摄像机以及所述进行后方拍摄的摄像机之间的位置距离，设置所述人车图片库中人车图片的存储时间；According to the position distance between the camera for shooting in front and the camera for shooting in the rear, the storage time of the pictures of people and cars in the picture library of people and cars is set;
当人车图片的存储时间超过所述设置的人车图片库中人车图片的存储时间时，将存储时间超过所述设置的存储时间的人车图片出库。When the storage time of the pictures of people and vehicles exceeds the storage time of the pictures of people and vehicles in the set picture library of people and vehicles, the pictures of people and vehicles whose storage time exceeds the set storage time are taken out of the warehouse.
可以理解的是，图5仅仅示出了骑行者重识别设备的简化设计。在实际应用中，电子设备还可以分别包含必要的其他元件，包含但不限于任意数量的输入/输出装置、处理器、控制器、存储器等，而所有可以实现本申请实施例的电子设备都在本申请的保护范围之内。It can be understood that Fig. 5 only shows a simplified design of the rider re-identification device. In practical applications, electronic equipment may also include other necessary components, including but not limited to any number of input/output devices, processors, controllers, memories, etc., and all electronic equipment that can implement the embodiments of the present application are in Within the protection scope of this application.
根据本发明实施例提供的一种骑行者重识别设备，通过对行人重识别模型进行迁移学习，得到骑行者重识别模型，利用该骑行者重识别模型提取人车图片库中多个人车图片的特征向量，计算特征向量之间的相似度值，并将最高的所述相似度值对应的至少两个所述人车图片识别为同一骑行者，提高了骑行者重识别的匹配度，实现对非机动车辆违法行为的监控；且根据抓拍时间和/或抓拍地点，选取目标人车图片库，提取目标人车图片库中的人车图片的特征向量进行相似度匹配，以提高人车图片的特征向量匹配的效率；且对存储超时的人车图片及时出库，以节省存储空间。According to a cyclist re-identification device provided in an embodiment of the present invention, the cyclist re-identification model is obtained by performing transfer learning on the pedestrian re-identification model, and the cyclist re-identification model is used to extract the images of multiple people and vehicles in the person-vehicle image database Eigenvectors, calculating the similarity value between the eigenvectors, and identifying at least two of the pictures of people and vehicles corresponding to the highest similarity value as the same cyclist, which improves the matching degree of re-identification of the cyclist, and realizes Monitoring of illegal behavior of non-motor vehicles; and according to the capture time and/or capture location, select the target person-vehicle image library, extract the feature vectors of the person-vehicle images in the target person-vehicle image library for similarity matching, in order to improve the accuracy of person-vehicle images. The efficiency of eigenvector matching; and timely release of pictures of people and vehicles that have been stored over time to save storage space.
所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的系统、装置和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。Those skilled in the art can clearly understand that for the convenience and brevity of the description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the foregoing method embodiment, which will not be repeated here.
在本申请所提供的几个实施例中，应该理解到，所揭露的系统、装置和方法，可以通过其它的方式实现。例如，该单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如，多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。所显示或讨论的相互之间的耦合、或直接耦合、或通信连接可以是通过一些接口，装置或单元的间接耦合或通信连接，可以是电性，机械或其它的形式。In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods may be implemented in other ways. For example, the division of this unit is only a logical function division, and there may be other division methods in actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored, or not implement. The mutual coupling, or direct coupling, or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be in electrical, mechanical or other forms.
作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。A unit described as a separate component may or may not be physically separated, and a component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
在上述实施例中，可以全部或部分地通过软件、硬件、固件或者其任意组合来实现。当使用软件实现时，可以全部或部分地以计算机程序产品的形式实现。该计算机程序产品包括一个或多个计算机指令。在计算机上加载和执行该计算机程序指令时，全部或部分地产生按照本申请实施例的流程或功能。该计算机可以是通用计算机、专用计算机、计算机网络、或者其他可编程装置。该计算机指令可以存储在计算机可读存储介质中，或者通过该计算机可读存储介质进行传输。该计算机指令可以从一个网站站点、计算机、服务器或数据中心通过有线(例如同轴电缆、光纤、数字用户线(digital subscriber line，DSL))或无线(例如红外、无线、微波等)方式向另一个网站站点、计算机、服务器或数据中心进行传输。该计算机可读存储介质可以是计算机能够存取的任何可用介质或者是包含一个或多个可用介质集成的服务器、数据中心等数据存储设备。该可用介质可以是只读存储器(read-onlymemory，ROM)，或随机存储存储器(random access memory，RAM)，或磁性介质，例如，软盘、硬盘、磁带、磁碟、或光介质，例如，数字通用光盘(digital versatile disc，DVD)、或者半导体介质，例如，固态硬盘(solid state disk，SSD)等。In the above embodiments, all or part of them may be implemented by software, hardware, firmware or any combination thereof. When implemented using software, it may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the processes or functions according to the embodiments of the present application will be generated in whole or in part. The computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable device. The computer instructions may be stored in or transmitted over a computer-readable storage medium. The computer instructions can be sent from one website site, computer, server, or data center to another via wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) A website site, computer, server or data center for transmission. The computer-readable storage medium may be any available medium that can be accessed by a computer, or a data storage device such as a server or a data center integrated with one or more available media. The available medium may be a read-only memory (ROM), or a random access memory (RAM), or a magnetic medium, such as a floppy disk, a hard disk, a magnetic tape, a magnetic disk, or an optical medium, such as a digital A universal optical disk (digital versatile disc, DVD), or a semiconductor medium, for example, a solid state disk (solid state disk, SSD) and the like.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910550548.6ACN110427814A (en) | 2019-06-24 | 2019-06-24 | A kind of bicyclist recognition methods, device and equipment again |
| PCT/CN2019/121517WO2020258714A1 (en) | 2019-06-24 | 2019-11-28 | Rider re-identification method, apparatus and device |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910550548.6ACN110427814A (en) | 2019-06-24 | 2019-06-24 | A kind of bicyclist recognition methods, device and equipment again |
| Publication Number | Publication Date |
|---|---|
| CN110427814Atrue CN110427814A (en) | 2019-11-08 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910550548.6APendingCN110427814A (en) | 2019-06-24 | 2019-06-24 | A kind of bicyclist recognition methods, device and equipment again |
| Country | Link |
|---|---|
| CN (1) | CN110427814A (en) |
| WO (1) | WO2020258714A1 (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111582277A (en)* | 2020-06-15 | 2020-08-25 | 深圳天海宸光科技有限公司 | License plate recognition system and method based on transfer learning |
| WO2020258714A1 (en)* | 2019-06-24 | 2020-12-30 | 深圳云天励飞技术有限公司 | Rider re-identification method, apparatus and device |
| CN112464922A (en)* | 2021-02-02 | 2021-03-09 | 长沙海信智能系统研究院有限公司 | Human-vehicle weight recognition and model training method, device, equipment and storage medium thereof |
| CN113129597A (en)* | 2019-12-31 | 2021-07-16 | 深圳云天励飞技术有限公司 | Method and device for identifying illegal vehicles on motor vehicle lane |
| CN113627352A (en)* | 2021-08-12 | 2021-11-09 | 塔里木大学 | A pedestrian re-identification method and system |
| CN114445787A (en)* | 2021-12-31 | 2022-05-06 | 深圳云天励飞技术股份有限公司 | Non-motor vehicle weight recognition method and related equipment |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114821629A (en)* | 2021-01-27 | 2022-07-29 | 天津大学 | Pedestrian re-identification method for performing cross image feature fusion based on neural network parallel training architecture |
| CN114863547A (en)* | 2022-03-22 | 2022-08-05 | 武汉众智数字技术有限公司 | A target detection method and system for removing cyclists |
| CN119249135B (en)* | 2024-12-06 | 2025-03-18 | 浙江大学 | A method and device for extracting aggressive riding features based on motorcycle running status |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107944017A (en)* | 2017-12-11 | 2018-04-20 | 浙江捷尚视觉科技股份有限公司 | The search method of non-motor vehicle in a kind of video |
| CN108021933A (en)* | 2017-11-23 | 2018-05-11 | 深圳市华尊科技股份有限公司 | Neural network recognization model and recognition methods |
| CN109344842A (en)* | 2018-08-15 | 2019-02-15 | 天津大学 | A Pedestrian Re-identification Method Based on Semantic Region Representation |
| CN109446898A (en)* | 2018-09-20 | 2019-03-08 | 暨南大学 | A kind of recognition methods again of the pedestrian based on transfer learning and Fusion Features |
| CN109657552A (en)* | 2018-11-16 | 2019-04-19 | 北京邮电大学 | The vehicle type recognition device being cold-started across scene and method are realized based on transfer learning |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106778804B (en)* | 2016-11-18 | 2020-10-20 | 天津大学 | A zero-shot image classification method based on class attribute transfer learning |
| CN109002761A (en)* | 2018-06-13 | 2018-12-14 | 中山大学新华学院 | A kind of pedestrian's weight identification monitoring system based on depth convolutional neural networks |
| CN109117823A (en)* | 2018-08-31 | 2019-01-01 | 常州大学 | A kind of across the scene pedestrian based on multilayer neural network knows method for distinguishing again |
| CN110427814A (en)* | 2019-06-24 | 2019-11-08 | 深圳云天励飞技术有限公司 | A kind of bicyclist recognition methods, device and equipment again |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108021933A (en)* | 2017-11-23 | 2018-05-11 | 深圳市华尊科技股份有限公司 | Neural network recognization model and recognition methods |
| CN107944017A (en)* | 2017-12-11 | 2018-04-20 | 浙江捷尚视觉科技股份有限公司 | The search method of non-motor vehicle in a kind of video |
| CN109344842A (en)* | 2018-08-15 | 2019-02-15 | 天津大学 | A Pedestrian Re-identification Method Based on Semantic Region Representation |
| CN109446898A (en)* | 2018-09-20 | 2019-03-08 | 暨南大学 | A kind of recognition methods again of the pedestrian based on transfer learning and Fusion Features |
| CN109657552A (en)* | 2018-11-16 | 2019-04-19 | 北京邮电大学 | The vehicle type recognition device being cold-started across scene and method are realized based on transfer learning |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020258714A1 (en)* | 2019-06-24 | 2020-12-30 | 深圳云天励飞技术有限公司 | Rider re-identification method, apparatus and device |
| CN113129597A (en)* | 2019-12-31 | 2021-07-16 | 深圳云天励飞技术有限公司 | Method and device for identifying illegal vehicles on motor vehicle lane |
| CN111582277A (en)* | 2020-06-15 | 2020-08-25 | 深圳天海宸光科技有限公司 | License plate recognition system and method based on transfer learning |
| CN112464922A (en)* | 2021-02-02 | 2021-03-09 | 长沙海信智能系统研究院有限公司 | Human-vehicle weight recognition and model training method, device, equipment and storage medium thereof |
| CN112464922B (en)* | 2021-02-02 | 2021-05-28 | 长沙海信智能系统研究院有限公司 | Human-vehicle weight recognition and model training method, device, equipment and storage medium thereof |
| CN113627352A (en)* | 2021-08-12 | 2021-11-09 | 塔里木大学 | A pedestrian re-identification method and system |
| CN114445787A (en)* | 2021-12-31 | 2022-05-06 | 深圳云天励飞技术股份有限公司 | Non-motor vehicle weight recognition method and related equipment |
| Publication number | Publication date |
|---|---|
| WO2020258714A1 (en) | 2020-12-30 |
| Publication | Publication Date | Title |
|---|---|---|
| CN110427814A (en) | A kind of bicyclist recognition methods, device and equipment again | |
| CN109784186B (en) | A pedestrian re-identification method, device, electronic device, and computer-readable storage medium | |
| CN111104867B (en) | Recognition model training and vehicle re-recognition method and device based on part segmentation | |
| CN111797653B (en) | Image labeling method and device based on high-dimensional image | |
| Abdullah et al. | YOLO-based three-stage network for Bangla license plate recognition in Dhaka metropolitan city | |
| CN109740573B (en) | Video analysis method, device, equipment and server | |
| WO2022134387A1 (en) | Vehicle wrong-way travel detection method, apparatus, device, computer-readable storage medium, and computer program product | |
| EP3096292A1 (en) | Multi-object tracking with generic object proposals | |
| CN111400550A (en) | A target motion trajectory construction method, device and computer storage medium | |
| Razalli et al. | Emergency vehicle recognition and classification method using HSV color segmentation | |
| WO2020155790A1 (en) | Method and apparatus for extracting claim settlement information, and electronic device | |
| CN110111565A (en) | A kind of people's vehicle flowrate System and method for flowed down based on real-time video | |
| CN112949751A (en) | Vehicle image clustering and track restoring method | |
| CN112614102A (en) | Vehicle detection method, terminal and computer readable storage medium thereof | |
| CN111753766B (en) | Image processing method, device, equipment and medium | |
| CN111753601B (en) | Image processing method, device and storage medium | |
| CN112699769A (en) | Detection method and system for left-over articles in security monitoring | |
| WO2022228325A1 (en) | Behavior detection method, electronic device, and computer readable storage medium | |
| CN103745223A (en) | Face detection method and apparatus | |
| CN108171135A (en) | Method for detecting human face, device and computer readable storage medium | |
| CN115730097A (en) | Human face filing method, device, equipment and medium based on personnel re-identification | |
| CN116052059A (en) | Traffic illegal behavior detection method, device and system | |
| CN114926973B (en) | Video monitoring method, device, system, server and readable storage medium | |
| CN115131725A (en) | Traffic flow statistical method, device, equipment and storage medium | |
| CN114677627A (en) | Target clue finding method, device, equipment and medium |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information | Address after:518000 1st floor, building 17, Shenzhen Dayun software Town, 8288 Longgang Avenue, Yuanshan street, Longgang District, Shenzhen City, Guangdong Province Applicant after:Shenzhen Yuntian lifeI Technology Co.,Ltd. Address before:518000 1st floor, building 17, Shenzhen Dayun software Town, 8288 Longgang Avenue, Yuanshan street, Longgang District, Shenzhen City, Guangdong Province Applicant before:SHENZHEN INTELLIFUSION TECHNOLOGIES Co.,Ltd. | |
| CB02 | Change of applicant information | ||
| RJ01 | Rejection of invention patent application after publication | Application publication date:20191108 | |
| RJ01 | Rejection of invention patent application after publication |