技术领域technical field
本发明属于图像处理技术领域，更进一步涉及基于图像的模式识别技术领域的一种基于卷积神经网络的车标识别方法。本发明针对交通系统中，由路口设置的高清拍照设备所得的车辆图片，进行汽车的车标定位，进而对汽车的车标进行识别，实现对车标的自动化定位并识别。The invention belongs to the technical field of image processing, and further relates to a vehicle logo recognition method based on a convolutional neural network in the technical field of image-based pattern recognition. In the traffic system, the invention uses the vehicle pictures obtained by the high-definition camera equipment installed at the intersection to locate the vehicle logo, and then recognize the vehicle logo, so as to realize the automatic positioning and recognition of the vehicle logo.
背景技术Background technique
随着社会经济水平的不断提高和车辆的普及，规模不断扩大的交通事业对更加智能化的技术和系统的需求更大，智能交通系统已经成为社会生活的热点问题。车辆识别系统作为智能交通系统的重要组成部分，在高速公路入口、停车场无人管理、违章车辆自动记录等领域都有这广泛的应用，它的实现具有很大的经济价值和现实意义。With the continuous improvement of the socio-economic level and the popularization of vehicles, the ever-expanding transportation industry has a greater demand for more intelligent technologies and systems, and intelligent transportation systems have become a hot issue in social life. As an important part of the intelligent transportation system, the vehicle identification system has a wide range of applications in the fields of highway entrance, unmanned parking lot management, and automatic recording of illegal vehicles. Its realization has great economic value and practical significance.
车标识别是车辆识别的一个重要方面。车标识别技术是指以数字图像或视频信号流为对象，通过图像处理与自动识别方法，获得机动车辆品牌信息的一种实用技术。车标识别系统包括车标的定位和车标识别二项关键技术。由于车标本身具有的多样性以及不同环境条件下的差异性等特点，加上人为拍摄获得的图片信息中车标的位置不确定性，因此找到一种优秀的车标定位和识别方法一个多学科交叉且富有挑战性的技术问题。Vehicle logo recognition is an important aspect of vehicle identification. Vehicle logo recognition technology refers to a practical technology that uses digital images or video signal streams as objects to obtain motor vehicle brand information through image processing and automatic recognition methods. The vehicle logo recognition system includes two key technologies of vehicle logo positioning and vehicle logo recognition. Due to the diversity of the car logo itself and the differences under different environmental conditions, coupled with the uncertainty of the position of the car logo in the image information obtained by artificial photography, it is necessary to find an excellent car logo positioning and recognition method. A multidisciplinary Intersecting and challenging technical issues.
现有的车标定位的方法，大多采用边缘检测和灰度直方图模板匹配的方法，由于车标小，这类方法容易受到背景环境的影响。已经有一部分车标识别的方法被提出，特别是目前使用较多的基于方向梯度直方图HOG特征和支持向量机SVM分类器的识别方法，大部分都是基于车牌和车标的相对位置确定车标位置，然后提取车标的方向梯度直方图HOG特征，利用支持向量机SVM训练成分类器进行车标识别。在车标识别中，方向梯度直方图HOG加支持向量机SVM算法由于采用了方向梯度直方图HOG特征，方向梯度直方图HOG描述子生成过程冗长，导致速度慢，实时性差，由于梯度的性质，该描述子对噪点相当敏感。现有的大部分车标识别算法，过程复杂，计算量太大，识别率不高，容易受到环境条件的影响，所以需要新的研究方法的提出。Most of the existing vehicle logo positioning methods use edge detection and gray histogram template matching methods. Due to the small size of the car logo, these methods are easily affected by the background environment. Some methods of vehicle identification recognition have been proposed, especially the recognition methods based on the HOG feature of the direction gradient histogram and the support vector machine SVM classifier are currently used, most of which are based on the relative position of the license plate and the vehicle logo to determine the vehicle logo. position, and then extract the HOG feature of the directional gradient histogram of the car logo, and use the support vector machine SVM to train it into a classifier for car logo recognition. In vehicle logo recognition, the directional gradient histogram HOG plus support vector machine SVM algorithm uses the directional gradient histogram HOG feature, and the directional gradient histogram HOG descriptor generation process is lengthy, resulting in slow speed and poor real-time performance. Due to the nature of the gradient, This descriptor is quite sensitive to noise. Most of the existing vehicle logo recognition algorithms are complex in process, large in calculation, low in recognition rate, and easily affected by environmental conditions, so new research methods are needed.
近年来，随着大数据、深度学习研究的发展，卷积神经网络CNN已成为当前语音分析和图像识别领域的研究热点，它的权值共享网络结构使之更类似于生物神经网络，降低了网络模型的复杂度，减少了权值的数量。该优点在网络的输入是多维图像时表现的更为明显，使图像可以直接作为网络的输入，避免了传统识别算法中复杂的特征提取和数据重建过程。卷积网络是为识别二维形状而特殊设计的一个多层感知器，这种网络结构对平移、比例缩放、倾斜或者共他形式的变形具有高度不变性。In recent years, with the development of big data and deep learning research, convolutional neural network (CNN) has become a research hotspot in the field of speech analysis and image recognition. Its weight sharing network structure makes it more similar to biological neural network, reducing the The complexity of the network model reduces the number of weights. This advantage is more obvious when the input of the network is a multi-dimensional image, so that the image can be directly used as the input of the network, avoiding the complicated feature extraction and data reconstruction process in the traditional recognition algorithm. The convolutional network is a multi-layer perceptron specially designed to recognize two-dimensional shapes. This network structure is highly invariant to translation, scaling, tilting, or other forms of deformation.
D.F.Llorca,R.Arroyo,M.A.Sotelo在其发表的论文“VehiclelogorecognitionintrafficimagesusingHOGfeaturesandSVM”(Proceedingsofthe16thInternationalIEEEAnnualConferenceonIntelligentTransportationSystems，2013)中提出了一套基于方向梯度直方图HOG和支持向量机SVM的车标识别的方法。该方法首先车牌定位，利用车标处于车牌正上方的先验知识，在车牌上方使用滑动窗口提前候选目标区域，然后提取候选区域的的方向梯度直方图HOG特征，最后利用支持向量机SVM训练的分类器进行车标分类。该方法存在的不足之处是，其一，由于该方法采用了方向梯度直方图HOG特征，方向梯度直方图HOG描述子生成过程冗长，导致速度慢，实时性差。其二，由于该方法梯度的性质，方向梯度直方图HOG描述子对噪点相当敏感，容易受到噪声的干扰。D.F.Llorca, R.Arroyo, M.A.Sotelo proposed a set of vehicle logo recognition methods based on directional gradient histogram HOG and support vector machine SVM in their paper "VehiclelogorecognitionintrafficimagesusingHOGfeaturesandSVM" (Proceedingofthe16thInternationalIEEEAnnualConferenceonIntelligentTransportationSystems, 2013). The method first locates the license plate, uses the prior knowledge that the vehicle logo is directly above the license plate, uses a sliding window above the license plate to advance the candidate target area, then extracts the HOG feature of the direction gradient histogram of the candidate area, and finally uses the support vector machine SVM training The classifier performs vehicle logo classification. The disadvantages of this method are: First, because the method uses the HOG feature of the histogram of oriented gradients, the generation process of the HOG descriptor of the oriented gradients histogram is lengthy, resulting in slow speed and poor real-time performance. Second, due to the nature of the gradient of the method, the HOG descriptor of the histogram of oriented gradients is quite sensitive to noise and is easily disturbed by noise.
佳都新太科技股份有限公司申请的专利“一种基于模式识别的车标自动定位与识别方法”(专利申请号：CN201410367377，公开号：CN104182728A)中提出了一种基于模式识别的车标自动定位与识别方法。该方法首先利用车牌检测技术，获取车牌的大小与位置，从而根据车牌与车标的相对位置，进行车标的初定位，其次利用基于哈尔Haar特征的强分类器Adaboost算法进行车标的二次定位，得到若干疑似车标的区域，再次利用基于方向梯度直方图HOG特征的支持向量机SVM算法对疑似车标区域进行筛选，选取具有最大置信度的区域作为车标定位结果，最后利用基于HOG特征的支持向量机SVM算法进行车标的识别。该方法存在的不足之处是，在定位流程中采用了基于哈尔Haar特征的强分类器Adaboost算法和基于方向梯度直方图HOG特征的支持向量机SVM算法，在车标识别流程中采用了基于方向梯度直方图HOG特征的支持向量机SVM算法，总共采用了三个分类器，大大增加了计算复杂度。而且HOG描述子生成过程耗时长，导致速度慢，实时性差。The patent "A method for automatic positioning and recognition of vehicle logos based on pattern recognition" applied by Jiadu Xintai Technology Co., Ltd. (patent application number: CN201410367377, publication number: CN104182728A) proposes a vehicle logo automatic positioning and recognition method based on pattern recognition. positioning and identification methods. The method first uses the license plate detection technology to obtain the size and position of the license plate, and then performs the initial positioning of the vehicle logo according to the relative position of the license plate and the car logo. A number of suspected car logo areas are obtained, and the support vector machine SVM algorithm based on the HOG feature of the direction gradient histogram is used to screen the suspected car logo area, and the area with the maximum confidence is selected as the car logo positioning result, and finally the support vector machine based on the HOG feature is used The vector machine SVM algorithm is used to identify the vehicle logo. The disadvantage of this method is that in the positioning process, the strong classifier Adaboost algorithm based on the Haar feature and the support vector machine SVM algorithm based on the HOG feature of the direction gradient histogram are used. The support vector machine (SVM) algorithm for the HOG feature of the directional gradient histogram uses a total of three classifiers, which greatly increases the computational complexity. Moreover, the HOG descriptor generation process takes a long time, resulting in slow speed and poor real-time performance.
上海交通大学申请的专利“车标自动识别方法及系统”(专利申请号：CN201310170528，公开号：CN103279738A)中提出一种车标自动识别方法，包括离线训练子系统和在线识别子系统。该方法根据密集尺度不变特征变换dense-SIFT和视觉词的相关性，将密集尺度不变特征变换dense-SIFT映射成所有视觉词表示，增加特征描述性。采用支持向量机训练车标分类器，实现车标识别。该方法存在的不足之处是，由于采用了密集尺度不变特征变换dense-SIFT特征算子，维数高，计算时间长，实时性差。The patent "automatic identification method and system of vehicle logo" (patent application number: CN201310170528, publication number: CN103279738A) filed by Shanghai Jiaotong University proposes an automatic identification method for vehicle logos, including an offline training subsystem and an online identification subsystem. According to the correlation between the dense scale invariant feature transform dense-SIFT and visual words, the method maps the dense scale invariant feature transform dense-SIFT to all visual word representations to increase feature descriptiveness. A support vector machine is used to train a car logo classifier to realize car logo recognition. The disadvantage of this method is that due to the dense scale invariant feature transformation dense-SIFT feature operator, the dimension is high, the calculation time is long, and the real-time performance is poor.
发明内容Contents of the invention
本发明的目的是针对上述现有技术存在的不足，提出了一种基于卷积神经网络的车标识别方法。本发明与现有技术中其他车标识别技术相比计算量小，准确度高，适应性强。The object of the present invention is to propose a vehicle logo recognition method based on a convolutional neural network for the deficiencies in the above-mentioned prior art. Compared with other vehicle logo recognition technologies in the prior art, the invention has small calculation amount, high accuracy and strong adaptability.
本发明实现的具体步骤包括如下：The concrete steps that the present invention realizes comprise as follows:
1.一种基于卷积神经网络的车标识别方法，包括如下步骤：1. A car logo recognition method based on a convolutional neural network, comprising the steps of:
(1)输入交通路口中高清拍照设备获取的车标待检测图片；(1) Input the picture to be detected of the car logo obtained by the high-definition camera equipment at the traffic intersection;
(2)车标定位：(2) Vehicle logo positioning:
(2a)对输入的车标待检测图片进行二值化操作，得到二值化后的车标图片；(2a) Carry out binarization operation on the input car logo picture to be detected, and obtain the car logo picture after binarization;
(2b)对二值化后的车标图片进行腐蚀和膨胀的形态学操作，得到联通区域；(2b) Corrosion and expansion morphological operations are performed on the binarized car logo image to obtain the connected area;
(2c)在联通区域中利用车牌联通区域的宽高比和矩形特征，筛选出车牌联通区域，得到车牌联通区域的左上坐标(x1，y1)和右下坐标(x2，y2)；(2c) Use the aspect ratio and rectangular features of the license plate Unicom area in the Unicom area to filter out the license plate Unicom area, and obtain the upper left coordinates (x1, y1) and lower right coordinates (x2, y2) of the license plate Unicom area;
(2d)在待检测车标图片中，通过滑窗截取车标区域图，滑窗底边以车牌上边沿为起始，沿着车牌的中央线往上滑动3次截取候选区域，得到车标区域图；(2d) In the picture of the car logo to be detected, intercept the car logo area map through the sliding window. The bottom edge of the sliding window starts from the upper edge of the license plate, and slides up three times along the central line of the license plate to intercept the candidate area to obtain the car logo area map;
(3)构建并训练卷积神经网络CNN：(3) Construct and train the convolutional neural network CNN:
(3a)构建含有7层的卷积神经网络CNN，7层依次是卷积层Conv1，池化层Pool2,卷积层Conv3，池化层Pool4，全连接层Fc5，全连接层Fc6，分类层Softmax7；(3a) Construct a convolutional neural network CNN with 7 layers. The 7 layers are convolutional layer Conv1, pooling layer Pool2, convolutional layer Conv3, pooling layer Pool4, fully connected layer Fc5, fully connected layer Fc6, classification layer Softmax7;
(3b)输入已标记并灰度化的车标区域样本图片，训练卷积神经网络CNN，直到输出层的损失函数J(θ)≤0.0001，得到车标识别的卷积神经网络CNN；(3b) Input the sample image of the marked and gray-scaled vehicle logo area, train the convolutional neural network CNN until the loss function J(θ) of the output layer≤0.0001, and obtain the convolutional neural network CNN for vehicle logo recognition;
(4)车标识别：(4) Vehicle logo recognition:
(4a)对车标区域图进行灰度化操作；(4a) Carry out the grayscale operation on the vehicle logo area map;
(4b)将灰度化后的车标区域图分辨率归一化至38×38像素大小，得到处理后的车标图；(4b) Normalize the resolution of the grayscaled vehicle logo area map to a size of 38×38 pixels to obtain the processed car logo map;
(4c)将处理后的车标图输入车标识别的卷积神经网络CNN，最终输出结果。(4c) Input the processed car logo image into the convolutional neural network CNN for car logo recognition, and finally output the result.
本发明与现有技术相比具有以下优点：Compared with the prior art, the present invention has the following advantages:
第一、由于本发明采用先定位车牌区域，然后在车牌正上区域用滑窗截取车标区域的车标定位方法，克服了现有技术中直接提取车标易受背景环境影响，不能被准确提取出来的问题，使得本发明能准确地从复杂背景环境中提取出车标区域。First, because the present invention adopts the vehicle logo positioning method of first locating the license plate area, and then intercepting the car logo area with a sliding window in the area directly above the license plate, it overcomes the fact that the direct extraction of the car logo in the prior art is easily affected by the background environment and cannot be accurately detected. The extracted problem enables the present invention to accurately extract the vehicle logo area from the complex background environment.
第二、由于本发明采用基于卷积神经网络CNN的车标识别方法，通过卷积神经网络CNN中多层的网络自学习特征，避免了传统识别算法中需要人工设计特征的过程，而且卷积神经网络CNN自学习的特征对环境变化具有更高的鲁棒性，使得本发明降低了计算量和复杂性，提高了识别率，对复杂背景具有更强的适应性。Second, because the present invention adopts the car logo recognition method based on the convolutional neural network CNN, through the multi-layer network self-learning feature in the convolutional neural network CNN, the process of manually designing features is avoided in the traditional recognition algorithm, and the convolution The self-learning feature of the neural network CNN has higher robustness to environmental changes, which reduces the calculation amount and complexity, improves the recognition rate, and has stronger adaptability to complex backgrounds.
附图说明Description of drawings
图1是本发明的流程图；Fig. 1 is a flow chart of the present invention;
图2是本发明车标定位的示意图；Fig. 2 is a schematic diagram of the location of the vehicle logo of the present invention;
图3是本发明卷积神经网络CNN结构图；Fig. 3 is a convolutional neural network (CNN) structure diagram of the present invention;
图4是本发明的部分标记的车标图。Fig. 4 is a vehicle icon diagram of a part of the mark of the present invention.
具体实施方式detailed description
下面结合附图对本发明做进一步的描述。The present invention will be further described below in conjunction with the accompanying drawings.
参照图1，本发明实现的具体步骤如下：With reference to Fig. 1, the concrete steps that the present invention realizes are as follows:
步骤1，输入交通路口中高清拍照设备拍下的待检测图片。Step 1, input the picture to be detected taken by the high-definition camera equipment at the traffic intersection.
待检测图片包含清晰可见的标记车牌和车标，标准车牌区域大小为180×60像素。The image to be detected contains clearly visible license plates and logos, and the standard license plate area size is 180×60 pixels.
步骤2，车标定位。Step 2, car logo positioning.
对输入的待检测图片进行二值化操作，得到二值化后的图片，具体操作是：Perform a binarization operation on the input image to be detected to obtain a binarized image. The specific operation is:
第一步，选取50个标记样本中车牌底色的三原色红绿蓝RGB值，统计三原色红绿蓝RGB值的均值；The first step is to select the three primary colors red, green and blue RGB values of the license plate background color in 50 mark samples, and count the mean value of the three primary colors red, green and blue RGB values;
第二步，按照下式，对输入的车标待检测图片进行二值化操作：In the second step, according to the following formula, the binarization operation is performed on the input vehicle logo image to be detected:
其中，qij表示二值化后的图片的像素点的灰度值，r0,g0,b0分别表示样本图片像素点的三原色红绿蓝均值，rij，gij,bij分别表示待检测图片像素点的三原色红绿蓝值,i,j分别表示图片像素点的行数和列数。Among them, qij represents the gray value of the pixel point of the binarized image, r0 , g0 , b0 represent the mean values of the three primary colors red, green and blue of the pixel point of the sample image respectively, rij, gij , bij represent The red, green, and blue values of the three primary colors of the picture pixel to be detected, i and j represent the number of rows and columns of the picture pixel, respectively.
对二值化后的图片进行腐蚀和膨胀的形态学操作，得到联通区域。在联通区域中利用车牌联通区域的宽高比和矩形特征即车牌的宽高比为3:1，车牌宽大于高，且与拍摄道路的水平线的倾斜角小于5度，筛选出车牌联通区域如图2(a)，得到车牌联通区域的左上坐标(x1，y1)和右下坐标(x2，y2)。Perform morphological operations of erosion and expansion on the binarized image to obtain connected areas. In the Unicom area, use the aspect ratio and rectangular features of the license plate Unicom area, that is, the aspect ratio of the license plate is 3:1, the width of the license plate is greater than the height, and the inclination angle with the horizontal line of the photographed road is less than 5 degrees, and the license plate Unicom area is screened out. Figure 2(a), get the upper left coordinates (x1, y1 ) and lower right coordinates (x2, y2 ) of the connected area of the license plate.
在待检测图片中，用滑窗截取车标区域图，滑窗的窗口为正方形，其边长为步长为其中，l表示滑窗窗口的边长，h表示滑窗窗口的步长，(x1，y1)表示车牌联通区域的左上坐标，(x2，y2)表示车牌联通区域的右下坐标。滑窗下边以车牌上边沿为起始，沿着车牌的中央线往上滑动3次截取候选区域，如图2(b)所示，最终输出的车标区域图如图2(c)。In the picture to be detected, use the sliding window to intercept the area map of the car logo. The window of the sliding window is a square, and its side length is Step size is Among them, l represents the side length of the sliding window window, h represents the step size of the sliding window window, (x1, y1 ) represents the upper left coordinate of the license plate communication area, (x2, y2 ) represents the lower right coordinate of the license plate communication area . The lower edge of the sliding window starts from the upper edge of the license plate, slides up three times along the central line of the license plate to intercept the candidate area, as shown in Figure 2(b), and the final output image of the license plate area is shown in Figure 2(c).
步骤3，构建并训练卷积神经网络。Step 3, build and train a convolutional neural network.
如图3所示构建含有7层的卷积神经网络CNN，7层依次是卷积层Conv1，池化层Pool2,卷积层Conv3，池化层Pool4，全连接层Fc5，全连接层Fc6，分类层Softmax7，构建含有7层的车标识别的卷积神经网络CNN具体步骤如下：As shown in Figure 3, a convolutional neural network CNN with 7 layers is constructed. The 7 layers are the convolutional layer Conv1, the pooling layer Pool2, the convolutional layer Conv3, the pooling layer Pool4, the fully connected layer Fc5, and the fully connected layer Fc6. The classification layer is Softmax7, and the specific steps of constructing a convolutional neural network CNN with 7 layers of vehicle identification are as follows:
第1步，将38×38像素大小的车标区域图输入卷积层Conv1，对其进行块大小为5×5像素和步长为1像素的卷积操作，总共用32个卷积核，得到32张34×34像素大小的特征图；Step 1: Input the 38×38 pixel area map of the car logo into the convolutional layer Conv1, and perform a convolution operation with a block size of 5×5 pixels and a step size of 1 pixel. A total of 32 convolution kernels are used. Get 32 feature maps of 34×34 pixel size;
第2步，将卷积层Conv1输出的32张特征图输入到池化层Pool2，对其进行最大池化操作，池化块的大小为2×2像素，步长为1像素，得到32张分辨率为17×17像素的特征图；In the second step, the 32 feature maps output by the convolutional layer Conv1 are input to the pooling layer Pool2, and the maximum pooling operation is performed on them. The size of the pooling block is 2×2 pixels, and the step size is 1 pixel, and 32 feature maps are obtained. A feature map with a resolution of 17×17 pixels;
第3步，将池化层Pool2输出的32张特征图输入卷积层Conv3，对其进行块大小为5×5像素和步长为1像素的卷积操作，总共用64个卷积核，得到64张分辨率为13×13像素的特征图；In the third step, the 32 feature maps output by the pooling layer Pool2 are input into the convolutional layer Conv3, and the convolution operation with a block size of 5×5 pixels and a step size of 1 pixel is performed on it, and a total of 64 convolution kernels are used. Get 64 feature maps with a resolution of 13×13 pixels;
第4步，将卷积层Conv3输出的64张特征图输入池化层Pool4，对其进行最大池化操作，池化块的大小为2×2像素，步长为1像素，得到64张分辨率为7×7像素的特征图；Step 4: Input the 64 feature maps output by the convolutional layer Conv3 into the pooling layer Pool4, and perform the maximum pooling operation on it. The size of the pooling block is 2×2 pixels, and the step size is 1 pixel, and 64 resolution images are obtained. A feature map with a rate of 7×7 pixels;
第5步，将池化层Pool4输出的64张特征图输入全连接层Fc5，按照下式，对其中每一个像素点进行激活，得到激活后的特征图的像素点的值，将激活后的特征图以列的顺序排列成1维向量，得到1×3136维的特征向量：Step 5, input the 64 feature maps output by the pooling layer Pool4 into the fully connected layer Fc5, activate each pixel according to the following formula, obtain the pixel value of the activated feature map, and convert the activated The feature maps are arranged into 1-dimensional vectors in the order of columns, resulting in a 1×3136-dimensional feature vector:
其中，f(x)表示激活后的特征图的像素点的值，x表示激活前特征图的像素点的值，e表示一个无限不循环的自然常数，取值为2.7182；Among them, f(x) represents the value of the pixel point of the activated feature map, x represents the value of the pixel point of the feature map before activation, and e represents an infinite non-cyclic natural constant with a value of 2.7182;
第6步，将全连接层Fc5输出的特征向量输入全连接层Fc6，构成一般神经网络，输出为1×500维的特征向量；Step 6: Input the feature vector output by the fully connected layer Fc5 into the fully connected layer Fc6 to form a general neural network, and the output is a 1×500-dimensional feature vector;
第7步，将全连接层Fc6输出的特征向量输入分类层Softmax7，得到车标区域图的分类标签，该层会计算出每种分类标签的概率，并将最大概率的标签输出，其中softmax分类的期望函数如下：Step 7: Input the feature vector output by the fully connected layer Fc6 into the classification layer Softmax7 to obtain the classification labels of the vehicle logo area map. This layer will calculate the probability of each classification label and output the label with the highest probability, among which the softmax classification is The expected function is as follows:
其中，hθ(α)表示softmax分类的期望函数，α表示卷积神经网络CNN中全连接层Fc6输出的特征向量，β表示与卷积神经网络CNN中全连接层Fc6输出的特征向量α相对应的标签，p(β＝t|α；θ)表示输入为卷积神经网络CNN中全连接层Fc6输出的特征向量α时标签β等于t的概率，t∈1,2,...,k，θ表示模型参数且θ1,θ2,...,θk∈Rn+1，softmax分类损失函数如下：Among them, hθ (α) represents the expectation function of softmax classification, α represents the feature vector output by the fully connected layer Fc6 in the convolutional neural network CNN, and β represents the feature vector α output from the fully connected layer Fc6 in the convolutional neural network CNN. The corresponding label, p(β=t|α; θ) represents the probability that the label β is equal to t when the input is the feature vector α output by the fully connected layer Fc6 in the convolutional neural network CNN, t∈1,2,..., k, θ represents the model parameters and θ1 , θ2 ,...,θk ∈ Rn+1 , the softmax classification loss function is as follows:
其中，J(θ)表示损失函数，m表示车标区域图样本的数量，hθ(α)表示softmax分类的期望函数，α表示卷积神经网络CNN中全连接层Fc6输出的特征向量，β表示与卷积神经网络CNN中全连接层Fc6输出的特征向量α相对应的标签，θ表示模型参数。Among them, J(θ) represents the loss function, m represents the number of car logo area map samples, hθ (α) represents the expectation function of softmax classification, α represents the feature vector output by the fully connected layer Fc6 in the convolutional neural network CNN, β Indicates the label corresponding to the feature vector α output by the fully connected layer Fc6 in the convolutional neural network CNN, and θ indicates the model parameters.
输入已标记并灰度化的车标区域样本图片，训练卷积神经网络CNN，训练的过程如下：Input the sample image of the marked and grayscaled car logo area, and train the convolutional neural network CNN. The training process is as follows:
第一步，向前传播阶段，从样本集中取一个样本，将样本输入卷积神经网络CNN计算相应的实际输出，在此阶段，信息从输入层经过逐级的变换，传送到输出层；The first step is the forward propagation stage. A sample is taken from the sample set, and the sample is input into the convolutional neural network CNN to calculate the corresponding actual output. In this stage, the information is transformed step by step from the input layer and transmitted to the output layer;
第二步，向后传播阶段，计算实际输出与样本标签对应的理想输出的差，按极小化误差的方法，反向传播调整权矩阵；The second step, the backward propagation stage, calculates the difference between the actual output and the ideal output corresponding to the sample label, and adjusts the weight matrix by back propagation according to the method of minimizing the error;
第三步，重复第一步和第二步的操作，直到卷积神经网络CNN分类层Softmax7之后的输出的损失函数J(θ)≤0.0001为止，得到车标识别的卷积神经网络CNN。The third step is to repeat the operations of the first step and the second step until the loss function J(θ)≤0.0001 of the output after the convolutional neural network CNN classification layer Softmax7, and obtain the convolutional neural network CNN for vehicle identification.
步骤4，车标识别。Step 4, car logo recognition.
对车标区域图，进行灰度化操作，并将车标区域图分辨率缩小至38×38像素大小，得到处理后的车标图，如图4所示。将处理后的车标图输入车标识别的卷积神经网络CNN，最终输出结果。Carry out the grayscale operation on the vehicle logo area map, and reduce the resolution of the car logo area map to a size of 38×38 pixels, and obtain the processed car logo map, as shown in Figure 4. Input the processed car logo image into the convolutional neural network CNN for car logo recognition, and finally output the result.
下面结合仿真实验对本发明的效果做进一步的描述。The effects of the present invention will be further described below in combination with simulation experiments.
1、仿真实验条件：1. Simulation experiment conditions:
本发明所用的数据库为收集并制作的一组包含10类车标的车标库，其中每个车标有3600张训练图和240张测试图，共36000张训练图和2400张测试图，负样本为12万，包括各种非车标的图片。硬件平台为：IntelCore2DuoCPUE65502.33GHZ、3GBRAM，软件平台：vs2010,MATLABR2012a。The database used in the present invention is a collection and production of a group of car logos containing 10 types of car logos, wherein each car logo has 3600 training images and 240 testing images, a total of 36000 training images and 2400 testing images, negative samples 120,000, including pictures of various non-vehicle logos. The hardware platform is: IntelCore2DuoCPUE65502.33GHZ, 3GBRAM, the software platform: vs2010, MATLABR2012a.
2、实验内容与结果：2. Experimental content and results:
本发明首先收集车标数据库中的十种车标，共36000张训练数据和2400张测试数据。通过车牌定位，车标初定位和车标精确定位最终精确确定车标的区域，将次车标区域图输入到已经训练好的卷积神经网络CNN分类器中分类，得出结果。对照下表的仿真结果，在测试的2400张数据中，各个类别奥迪，本田，比亚迪，标志，别克，大众，丰田，Jeep，起亚，长安，分别错误个数为6张，10张，12张，5张，6张，8张,11张,7张,8张,9张。可以看出本发明在识别车标时具有较高的识别率。The present invention first collects ten kinds of car logos in the car logo database, a total of 36,000 pieces of training data and 2,400 pieces of test data. Through the license plate positioning, initial positioning of the car logo and precise positioning of the car logo, the area of the car logo is finally accurately determined, and the sub-vehicle logo area map is input into the trained convolutional neural network CNN classifier for classification, and the result is obtained. Comparing the simulation results in the table below, among the 2400 test data, the number of errors in each category of Audi, Honda, BYD, Peugeot, Buick, Volkswagen, Toyota, Jeep, Kia, and Changan is 6, 10, and 12, respectively. , 5 sheets, 6 sheets, 8 sheets, 11 sheets, 7 sheets, 8 sheets, 9 sheets. It can be seen that the present invention has a higher recognition rate when recognizing vehicle logos.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510523632.0ACN105354568A (en) | 2015-08-24 | 2015-08-24 | Convolutional neural network based vehicle logo identification method |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510523632.0ACN105354568A (en) | 2015-08-24 | 2015-08-24 | Convolutional neural network based vehicle logo identification method |
| Publication Number | Publication Date |
|---|---|
| CN105354568Atrue CN105354568A (en) | 2016-02-24 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510523632.0APendingCN105354568A (en) | 2015-08-24 | 2015-08-24 | Convolutional neural network based vehicle logo identification method |
| Country | Link |
|---|---|
| CN (1) | CN105354568A (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105678344A (en)* | 2016-02-29 | 2016-06-15 | 浙江群力电气有限公司 | Intelligent classification method for power instrument equipment |
| CN105868785A (en)* | 2016-03-30 | 2016-08-17 | 乐视控股（北京）有限公司 | Image identification method based on convolutional neural network and image identification system thereof |
| CN105957238A (en)* | 2016-05-20 | 2016-09-21 | 聚龙股份有限公司 | Banknote management method and system |
| CN105956524A (en)* | 2016-04-22 | 2016-09-21 | 北京智芯原动科技有限公司 | Method and device for identifying traffic signs |
| CN105975941A (en)* | 2016-05-31 | 2016-09-28 | 电子科技大学 | Multidirectional vehicle model detection recognition system based on deep learning |
| CN106022285A (en)* | 2016-05-30 | 2016-10-12 | 北京智芯原动科技有限公司 | Vehicle type identification method and vehicle type identification device based on convolutional neural network |
| CN106056751A (en)* | 2016-05-20 | 2016-10-26 | 聚龙股份有限公司 | Prefix number identification method and system |
| CN106156744A (en)* | 2016-07-11 | 2016-11-23 | 西安电子科技大学 | SAR target detection method based on CFAR detection with degree of depth study |
| CN106372402A (en)* | 2016-08-30 | 2017-02-01 | 中国石油大学(华东) | A Parallelization Method of Fuzzy Region Convolutional Neural Network in Big Data Environment |
| CN106372666A (en)* | 2016-08-31 | 2017-02-01 | 同观科技（深圳）有限公司 | Target identification method and device |
| CN106446514A (en)* | 2016-08-31 | 2017-02-22 | 中国石油大学(华东) | A logging facies identification method based on fuzzy theory and neural network |
| CN106443598A (en)* | 2016-12-08 | 2017-02-22 | 中国人民解放军海军航空工程学院 | Convolutional neural network based cooperative radar network track deception jamming discrimination method |
| CN106503710A (en)* | 2016-10-26 | 2017-03-15 | 北京邮电大学 | A kind of automobile logo identification method and device |
| CN106557768A (en)* | 2016-11-25 | 2017-04-05 | 北京小米移动软件有限公司 | The method and device is identified by word in picture |
| CN106651774A (en)* | 2016-12-27 | 2017-05-10 | 深圳市捷顺科技实业股份有限公司 | License plate super-resolution model reconstruction method and device |
| CN106934392A (en)* | 2017-02-28 | 2017-07-07 | 西交利物浦大学 | Vehicle-logo recognition and attribute forecast method based on multi-task learning convolutional neural networks |
| CN106991420A (en)* | 2017-03-27 | 2017-07-28 | 新智认知数据服务有限公司 | A kind of detection method of license plate of the license plate area regression technique based on piecemeal |
| CN107134144A (en)* | 2017-04-27 | 2017-09-05 | 武汉理工大学 | A kind of vehicle checking method for traffic monitoring |
| CN107153873A (en)* | 2017-05-08 | 2017-09-12 | 中国科学院计算技术研究所 | A kind of two-value convolutional neural networks processor and its application method |
| CN107169402A (en)* | 2016-03-08 | 2017-09-15 | 福特全球技术公司 | Vehicle lane is positioned |
| CN107590492A (en)* | 2017-08-28 | 2018-01-16 | 浙江工业大学 | A kind of vehicle-logo location and recognition methods based on convolutional neural networks |
| CN107590456A (en)* | 2017-09-06 | 2018-01-16 | 张栖瀚 | Small micro- mesh object detection method in a kind of high-altitude video monitoring |
| CN107871125A (en)* | 2017-11-14 | 2018-04-03 | 深圳码隆科技有限公司 | Architecture against regulations recognition methods, device and electronic equipment |
| CN108156130A (en)* | 2017-03-27 | 2018-06-12 | 上海观安信息技术股份有限公司 | Network attack detecting method and device |
| CN108268860A (en)* | 2018-02-09 | 2018-07-10 | 重庆科技学院 | A kind of gas gathering and transportation station equipment image classification method based on convolutional neural networks |
| CN108403111A (en)* | 2018-02-01 | 2018-08-17 | 华中科技大学 | A kind of epileptic electroencephalogram (eeg) recognition methods and system based on convolutional neural networks |
| US10157441B2 (en) | 2016-12-27 | 2018-12-18 | Automotive Research & Testing Center | Hierarchical system for detecting object with parallel architecture and hierarchical method thereof |
| CN109389014A (en)* | 2017-08-10 | 2019-02-26 | 杭州海康威视数字技术股份有限公司 | Apply detection method, device and the electronic equipment of license plate vehicle |
| CN105809704B (en)* | 2016-03-30 | 2019-03-15 | 北京小米移动软件有限公司 | Method and device for identifying image clarity |
| CN109492681A (en)* | 2018-10-26 | 2019-03-19 | 北京交通大学 | Automobile logo identification method based on compound CNN model |
| CN109606678A (en)* | 2018-11-22 | 2019-04-12 | 东南大学 | A crawler-type unmanned aerial vehicle for automatic positioning of bridge supports |
| CN109747681A (en)* | 2019-01-14 | 2019-05-14 | 卡斯柯信号有限公司 | A train positioning device and method |
| CN110084215A (en)* | 2019-05-05 | 2019-08-02 | 上海海事大学 | A kind of pedestrian of the twin network model of binaryzation triple recognition methods and system again |
| CN110261749A (en)* | 2019-07-24 | 2019-09-20 | 广东电网有限责任公司 | A kind of GIS partial discharge fault identification model building method, device and fault recognition method |
| CN110569921A (en)* | 2019-09-17 | 2019-12-13 | 中控智慧科技股份有限公司 | Vehicle logo identification method, system, device and computer readable medium |
| CN110619304A (en)* | 2019-09-17 | 2019-12-27 | 中控智慧科技股份有限公司 | Vehicle type recognition method, system, device and computer readable medium |
| CN110633712A (en)* | 2019-09-17 | 2019-12-31 | 中控智慧科技股份有限公司 | Method, system and device for recognizing vehicle body color and computer readable medium |
| CN110765862A (en)* | 2019-09-17 | 2020-02-07 | 中控智慧科技股份有限公司 | Car logo identification method and terminal equipment |
| CN110889318A (en)* | 2018-09-05 | 2020-03-17 | 斯特拉德视觉公司 | Lane detection method and apparatus using CNN |
| CN110969135A (en)* | 2019-12-05 | 2020-04-07 | 中南大学 | Vehicle logo recognition method in natural scene |
| CN111091150A (en)* | 2019-12-12 | 2020-05-01 | 哈尔滨市科佳通用机电股份有限公司 | Railway wagon cross rod cover plate fracture detection method |
| US11049007B2 (en) | 2016-05-06 | 2021-06-29 | Fujitsu Limited | Recognition apparatus based on deep neural network, training apparatus and methods thereof |
| CN116310999A (en)* | 2023-05-05 | 2023-06-23 | 贵州中水能源股份有限公司 | Method for detecting large floaters in reservoir area of hydroelectric power station |
| US12272131B2 (en) | 2016-11-07 | 2025-04-08 | The Nielsen Company (Us), Llc | Recurrent deep neural network system for detecting overlays in images |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090208058A1 (en)* | 2004-04-15 | 2009-08-20 | Donnelly Corporation | Imaging system for vehicle |
| CN104134067A (en)* | 2014-07-07 | 2014-11-05 | 河海大学常州校区 | Road vehicle monitoring system based on intelligent visual Internet of Things |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090208058A1 (en)* | 2004-04-15 | 2009-08-20 | Donnelly Corporation | Imaging system for vehicle |
| CN104134067A (en)* | 2014-07-07 | 2014-11-05 | 河海大学常州校区 | Road vehicle monitoring system based on intelligent visual Internet of Things |
| Title |
|---|
| D.F.LLORCA等: "Vehicle logo recognition in traffic images using HOG features and SVM", 《IEEE》* |
| YANN LECUN等: "Gradient-based learning applied to document recognition", 《IEEE》* |
| 卢雅琴等: "基于数学形态学的车牌定位方法", 《计算机工程》* |
| 彭博: "基于深度学习的车标识别方法研究", 《计算机科学》* |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105678344A (en)* | 2016-02-29 | 2016-06-15 | 浙江群力电气有限公司 | Intelligent classification method for power instrument equipment |
| CN107169402B (en)* | 2016-03-08 | 2021-10-08 | 福特全球技术公司 | vehicle lane positioning |
| CN107169402A (en)* | 2016-03-08 | 2017-09-15 | 福特全球技术公司 | Vehicle lane is positioned |
| CN105809704B (en)* | 2016-03-30 | 2019-03-15 | 北京小米移动软件有限公司 | Method and device for identifying image clarity |
| CN105868785A (en)* | 2016-03-30 | 2016-08-17 | 乐视控股（北京）有限公司 | Image identification method based on convolutional neural network and image identification system thereof |
| WO2017166586A1 (en)* | 2016-03-30 | 2017-10-05 | 乐视控股（北京）有限公司 | Image identification method and system based on convolutional neural network, and electronic device |
| CN105956524A (en)* | 2016-04-22 | 2016-09-21 | 北京智芯原动科技有限公司 | Method and device for identifying traffic signs |
| US11049007B2 (en) | 2016-05-06 | 2021-06-29 | Fujitsu Limited | Recognition apparatus based on deep neural network, training apparatus and methods thereof |
| CN106056751A (en)* | 2016-05-20 | 2016-10-26 | 聚龙股份有限公司 | Prefix number identification method and system |
| CN106056751B (en)* | 2016-05-20 | 2019-04-12 | 聚龙股份有限公司 | The recognition methods and system of serial number |
| US10930105B2 (en) | 2016-05-20 | 2021-02-23 | Julong Co., Ltd. | Banknote management method and system |
| CN105957238B (en)* | 2016-05-20 | 2019-02-19 | 聚龙股份有限公司 | A kind of paper currency management method and its system |
| CN105957238A (en)* | 2016-05-20 | 2016-09-21 | 聚龙股份有限公司 | Banknote management method and system |
| CN106022285A (en)* | 2016-05-30 | 2016-10-12 | 北京智芯原动科技有限公司 | Vehicle type identification method and vehicle type identification device based on convolutional neural network |
| CN105975941A (en)* | 2016-05-31 | 2016-09-28 | 电子科技大学 | Multidirectional vehicle model detection recognition system based on deep learning |
| CN106156744A (en)* | 2016-07-11 | 2016-11-23 | 西安电子科技大学 | SAR target detection method based on CFAR detection with degree of depth study |
| CN106156744B (en)* | 2016-07-11 | 2019-01-29 | 西安电子科技大学 | SAR target detection method based on CFAR detection and deep learning |
| CN106372402B (en)* | 2016-08-30 | 2019-04-30 | 中国石油大学(华东) | Parallelization method of fuzzy area convolution neural network in big data environment |
| CN106372402A (en)* | 2016-08-30 | 2017-02-01 | 中国石油大学(华东) | A Parallelization Method of Fuzzy Region Convolutional Neural Network in Big Data Environment |
| CN106446514A (en)* | 2016-08-31 | 2017-02-22 | 中国石油大学(华东) | A logging facies identification method based on fuzzy theory and neural network |
| CN106372666B (en)* | 2016-08-31 | 2019-07-19 | 同观科技（深圳）有限公司 | A kind of target identification method and device |
| CN106372666A (en)* | 2016-08-31 | 2017-02-01 | 同观科技（深圳）有限公司 | Target identification method and device |
| CN106503710A (en)* | 2016-10-26 | 2017-03-15 | 北京邮电大学 | A kind of automobile logo identification method and device |
| US12272131B2 (en) | 2016-11-07 | 2025-04-08 | The Nielsen Company (Us), Llc | Recurrent deep neural network system for detecting overlays in images |
| CN106557768A (en)* | 2016-11-25 | 2017-04-05 | 北京小米移动软件有限公司 | The method and device is identified by word in picture |
| CN106443598A (en)* | 2016-12-08 | 2017-02-22 | 中国人民解放军海军航空工程学院 | Convolutional neural network based cooperative radar network track deception jamming discrimination method |
| CN106651774A (en)* | 2016-12-27 | 2017-05-10 | 深圳市捷顺科技实业股份有限公司 | License plate super-resolution model reconstruction method and device |
| US10157441B2 (en) | 2016-12-27 | 2018-12-18 | Automotive Research & Testing Center | Hierarchical system for detecting object with parallel architecture and hierarchical method thereof |
| CN106934392A (en)* | 2017-02-28 | 2017-07-07 | 西交利物浦大学 | Vehicle-logo recognition and attribute forecast method based on multi-task learning convolutional neural networks |
| CN106934392B (en)* | 2017-02-28 | 2020-05-26 | 西交利物浦大学 | Vehicle logo recognition and attribute prediction method based on multi-task learning convolutional neural network |
| CN106991420A (en)* | 2017-03-27 | 2017-07-28 | 新智认知数据服务有限公司 | A kind of detection method of license plate of the license plate area regression technique based on piecemeal |
| CN108156130A (en)* | 2017-03-27 | 2018-06-12 | 上海观安信息技术股份有限公司 | Network attack detecting method and device |
| CN107134144A (en)* | 2017-04-27 | 2017-09-05 | 武汉理工大学 | A kind of vehicle checking method for traffic monitoring |
| CN107153873B (en)* | 2017-05-08 | 2018-06-01 | 中国科学院计算技术研究所 | A kind of two-value convolutional neural networks processor and its application method |
| CN107153873A (en)* | 2017-05-08 | 2017-09-12 | 中国科学院计算技术研究所 | A kind of two-value convolutional neural networks processor and its application method |
| CN109389014A (en)* | 2017-08-10 | 2019-02-26 | 杭州海康威视数字技术股份有限公司 | Apply detection method, device and the electronic equipment of license plate vehicle |
| CN109389014B (en)* | 2017-08-10 | 2021-05-28 | 杭州海康威视数字技术股份有限公司 | Detection method and device for license plate-applied vehicle and electronic equipment |
| CN107590492A (en)* | 2017-08-28 | 2018-01-16 | 浙江工业大学 | A kind of vehicle-logo location and recognition methods based on convolutional neural networks |
| CN107590492B (en)* | 2017-08-28 | 2019-11-19 | 浙江工业大学 | A method of vehicle logo location and recognition based on convolutional neural network |
| CN107590456B (en)* | 2017-09-06 | 2020-09-18 | 张栖瀚 | Method for detecting small and micro targets in high-altitude video monitoring |
| CN107590456A (en)* | 2017-09-06 | 2018-01-16 | 张栖瀚 | Small micro- mesh object detection method in a kind of high-altitude video monitoring |
| CN107871125A (en)* | 2017-11-14 | 2018-04-03 | 深圳码隆科技有限公司 | Architecture against regulations recognition methods, device and electronic equipment |
| CN108403111A (en)* | 2018-02-01 | 2018-08-17 | 华中科技大学 | A kind of epileptic electroencephalogram (eeg) recognition methods and system based on convolutional neural networks |
| CN108268860A (en)* | 2018-02-09 | 2018-07-10 | 重庆科技学院 | A kind of gas gathering and transportation station equipment image classification method based on convolutional neural networks |
| CN110889318B (en)* | 2018-09-05 | 2024-01-19 | 斯特拉德视觉公司 | Lane detection method and device using CNN |
| CN110889318A (en)* | 2018-09-05 | 2020-03-17 | 斯特拉德视觉公司 | Lane detection method and apparatus using CNN |
| CN109492681A (en)* | 2018-10-26 | 2019-03-19 | 北京交通大学 | Automobile logo identification method based on compound CNN model |
| CN109606678B (en)* | 2018-11-22 | 2021-09-07 | 东南大学 | A crawler-type unmanned aerial vehicle for automatic positioning of bridge supports |
| CN109606678A (en)* | 2018-11-22 | 2019-04-12 | 东南大学 | A crawler-type unmanned aerial vehicle for automatic positioning of bridge supports |
| CN109747681A (en)* | 2019-01-14 | 2019-05-14 | 卡斯柯信号有限公司 | A train positioning device and method |
| CN110084215A (en)* | 2019-05-05 | 2019-08-02 | 上海海事大学 | A kind of pedestrian of the twin network model of binaryzation triple recognition methods and system again |
| CN110261749A (en)* | 2019-07-24 | 2019-09-20 | 广东电网有限责任公司 | A kind of GIS partial discharge fault identification model building method, device and fault recognition method |
| CN110765862A (en)* | 2019-09-17 | 2020-02-07 | 中控智慧科技股份有限公司 | Car logo identification method and terminal equipment |
| CN110633712A (en)* | 2019-09-17 | 2019-12-31 | 中控智慧科技股份有限公司 | Method, system and device for recognizing vehicle body color and computer readable medium |
| CN110619304A (en)* | 2019-09-17 | 2019-12-27 | 中控智慧科技股份有限公司 | Vehicle type recognition method, system, device and computer readable medium |
| CN110569921A (en)* | 2019-09-17 | 2019-12-13 | 中控智慧科技股份有限公司 | Vehicle logo identification method, system, device and computer readable medium |
| CN110969135A (en)* | 2019-12-05 | 2020-04-07 | 中南大学 | Vehicle logo recognition method in natural scene |
| CN110969135B (en)* | 2019-12-05 | 2023-04-18 | 中南大学 | Vehicle logo recognition method in natural scene |
| CN111091150A (en)* | 2019-12-12 | 2020-05-01 | 哈尔滨市科佳通用机电股份有限公司 | Railway wagon cross rod cover plate fracture detection method |
| CN116310999A (en)* | 2023-05-05 | 2023-06-23 | 贵州中水能源股份有限公司 | Method for detecting large floaters in reservoir area of hydroelectric power station |
| CN116310999B (en)* | 2023-05-05 | 2023-07-21 | 贵州中水能源股份有限公司 | Method for detecting large floaters in reservoir area of hydroelectric power station |
| Publication | Publication Date | Title |
|---|---|---|
| CN105354568A (en) | Convolutional neural network based vehicle logo identification method | |
| CN105868774A (en) | Selective search and convolutional neural network based vehicle logo recognition method | |
| CN107563385B (en) | License plate character recognition method based on deep convolutional generative adversarial network | |
| CN112200186B (en) | Vehicle logo identification method based on improved YOLO_V3 model | |
| CN105354572B (en) | A kind of automatic license plate identification system based on simplified convolutional neural networks | |
| CN108154102B (en) | Road traffic sign identification method | |
| CN111160205B (en) | An end-to-end unified detection method for embedded multi-type targets in traffic scenes | |
| CN106650731B (en) | A Robust License Plate and Vehicle Logo Recognition Method | |
| CN101937508B (en) | License plate localization and identification method based on high-definition image | |
| CN102663377B (en) | Character recognition method based on template matching | |
| CN102509091B (en) | Airplane tail number recognition method | |
| CN105335702B (en) | A kind of bayonet model recognizing method based on statistical learning | |
| CN105389550B (en) | It is a kind of based on sparse guide and the remote sensing target detection method that significantly drives | |
| CN110348376A (en) | A kind of pedestrian's real-time detection method neural network based | |
| CN105868700A (en) | Vehicle type recognition and tracking method and system based on monitoring video | |
| CN111695514A (en) | Vehicle detection method in foggy days based on deep learning | |
| CN104809461B (en) | The licence plate recognition method and system of binding sequence image super-resolution rebuilding | |
| CN105005989B (en) | A kind of vehicle target dividing method under weak contrast | |
| CN108009518A (en) | A kind of stratification traffic mark recognition methods based on quick two points of convolutional neural networks | |
| CN101859382A (en) | A license plate detection and recognition method based on the maximum stable extremum region | |
| CN106709530A (en) | License plate recognition method based on video | |
| CN105989334B (en) | Road detection method based on monocular vision | |
| CN105160309A (en) | Three-lane detection method based on image morphological segmentation and region growing | |
| CN103324920A (en) | Method for automatically identifying vehicle type based on vehicle frontal image and template matching | |
| CN109726717A (en) | A vehicle comprehensive information detection system |
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication | Application publication date:20160224 |