技术领域technical field
本发明属于计算机视觉与模式分析技术领域，具体涉及一种利用四轴飞行器判别车辆方位的方法。The invention belongs to the technical field of computer vision and pattern analysis, and in particular relates to a method for judging vehicle orientation by using a four-axis aircraft.
背景技术Background technique
随着城市的快速发展，城市交通日益发达，人们拥有的私家车数量也逐渐增加。截至2012年年底，上海注册的机动车保有量已经突破262万辆，其中私家车高达140万辆，而2013年上海私家车的保有量增幅更达到了15%左右，比去年增加22万辆，这使得交通形势更为严峻。随之而来的是车辆管理难度的加大。通过对一些人流量较大的地段观察，有些车停放在公共场所门口，有些车停在行车道上。这些车辆严重影响了交通道路的畅通，甚至可能引发交通事故。由于这种现象的分散不定，停车管理和违章检测等需要耗费较大的人力和时间成本。而我国目前车辆管理工作主要由大量交警人力完成，不仅不能保证及时性也耗费了大量人力资源，基于无人机的车辆管理就可以高效完成这项任务。管理中必不可少的步骤就是对车牌信息的记录，而无人机拍摄车牌的关键在于它能否准确的判断自身与车辆的相对位置并对准车牌。With the rapid development of cities, urban traffic is increasingly developed, and the number of private cars owned by people is gradually increasing. By the end of 2012, the number of registered motor vehicles in Shanghai had exceeded 2.62 million, including 1.4 million private cars. In 2013, the number of private cars in Shanghai increased by about 15%, an increase of 220,000 compared with last year. This makes the traffic situation even more severe. Followed by the increasing difficulty of vehicle management. Through the observation of some areas with a large flow of people, some cars are parked at the entrance of public places, and some cars are parked on the driveway. These vehicles have seriously affected the smooth flow of traffic roads, and may even cause traffic accidents. Due to the dispersion of this phenomenon, parking management and violation detection require a lot of manpower and time costs. However, the current vehicle management work in our country is mainly completed by a large number of traffic police manpower, which not only cannot guarantee timeliness but also consumes a lot of human resources. Vehicle management based on drones can efficiently complete this task. An essential step in management is to record the license plate information, and the key to taking pictures of the license plate by the drone is whether it can accurately judge the relative position of itself and the vehicle and align the license plate.
车辆方位检测在车辆识别，车辆自动导航和智能交通系统等领域都有着重要的应用。传统的基于图像序列的车辆方位检测通过块匹配方法计算相邻帧车辆的相关性来估测车辆的方位。然而这种方法是针对固定摄像头和运动车辆的场景，与我们车辆静止无人机运动的场景不符。我们需要的是从单张图片识别车辆方位的方法。文献[1]通过训练基于HOG特征的二分类器决策树识别车辆的方位，但是只能识别固定的8个方位。文献[2]在聚类的框架下基于颜色和边缘特征对车辆的方位进行了识别，但是也不能得到连续的方位角。Vehicle orientation detection has important applications in the fields of vehicle identification, vehicle automatic navigation and intelligent transportation systems. The traditional vehicle orientation detection based on image sequence calculates the correlation of vehicles in adjacent frames by block matching method to estimate the vehicle orientation. However, this method is aimed at the scene of fixed cameras and moving vehicles, which does not match the scene of our vehicle stationary UAV movement. What we need is a way to recognize a vehicle's orientation from a single image. Literature [1] recognizes the orientation of the vehicle by training a two-classifier decision tree based on HOG features, but it can only identify 8 fixed orientations. Literature [2] recognizes the orientation of the vehicle based on the color and edge features under the framework of clustering, but it cannot obtain continuous azimuth angles.
为了更准确的控制无人机的飞行状态，需要找到一种车辆方位识别算法来得到车辆的方位角信息。In order to control the flight state of the UAV more accurately, it is necessary to find a vehicle orientation recognition algorithm to obtain the vehicle's azimuth angle information.
参考文献references
[1]RybskiPE,HuberD,MorrisDD,etal.VisualclassificationofcoarsevehicleorientationusingHistogramofOrientedGradientsfeatures[C]//IntelligentVehiclesSymposium(IV),2010IEEEIEEE,2010:921-928.[1] RybskiPE, HuberD, MorrisDD, et al. Visual classification of coarse vehicle orientation using Histogram of Oriented Gradients features [C]//Intelligent Vehicles Symposium (IV), 2010IEEEIEEE, 2010:921-928.
[2]WuJC,HsiehJW,ChenYS,etal.VehicleOrientationDetectionUsingVehicleColorandNormalizedCutClustering.[J].VehicleOrientationDetectionUsingVehicleColorandNormalizedCutClustering.-ResearchGate,2007:457-460.[2] WuJC, HsiehJW, ChenYS, etal.VehicleOrientationDetectionUsingVehicleColorandNormalizedCutClustering.[J].
[3]DalalN,TriggsB.Histogramsoforientedgradientsforhumandetection[C]//ComputerVisionandPatternRecognition,2005.CVPR2005.IEEEComputerSocietyConferenceonIEEE,2005:886-893vol.1.[3]DalalN, TriggsB.Histogramsoforientedgradientsforhumandetection[C]//ComputerVisionandPatternRecognition, 2005.CVPR2005.IEEEComputerSocietyConferenceonIEEE,2005:886-893vol.1.
[4]BurgesCJC.Atutorialonsupportvectormachinesforpatternrecognition[C]//DataMiningandKnowledgeDiscovery1998:121--167.。[4]BurgesCJC.Atutorialonsupportvectormachinesforpatternrecognition[C]//DataMiningandKnowledgeDiscovery1998:121--167.
发明内容Contents of the invention
本发明的目的在于提出一种利用四轴飞行器判别车辆方位的方法。The purpose of the present invention is to propose a method for judging the orientation of a vehicle using a quadcopter.
本发明提出的利用四轴飞行器判别车辆方位的方法，关键在于如何有效的检测到车辆、得到车辆的方位角并对无人机的行为做出指导。本发明方法的具体步骤为：The key to the method of using a quadcopter to determine the orientation of a vehicle proposed by the present invention is how to effectively detect the vehicle, obtain the azimuth angle of the vehicle, and guide the behavior of the drone. The concrete steps of the inventive method are:
步骤1、对无人机拍摄到的视频数据进行车辆识别；Step 1, Carrying out vehicle recognition on the video data captured by the drone;
步骤2、对识别到的车辆，检测其车轮轮廓形状；Step 2. For the identified vehicle, detect its wheel profile shape;
步骤3、以步骤2的结果为根据，根据两个车轮的相对位置，判断车辆与飞行器的相对方位，为飞行器自主对准车牌提供控制信息。Step 3. Based on the result of step 2, according to the relative positions of the two wheels, determine the relative orientation of the vehicle and the aircraft, and provide control information for the aircraft to autonomously align the license plate.
下面对以上三个步骤进行具体描述。The above three steps are described in detail below.
步骤1、对无人机拍摄到的视频数据进行车辆识别Step 1. Carry out vehicle recognition on the video data captured by the drone
车辆识别采用基于HOG特征[3]的SVM[4]分类算法实现。主要分为HOG特征提取、SVM训练和SVM检测几个步骤，具体介绍如下：Vehicle recognition is implemented using the SVM[4] classification algorithm based on HOG features[3] . It is mainly divided into several steps of HOG feature extraction, SVM training and SVM detection. The details are as follows:
（1）HOG特征提取(1) HOG feature extraction
HOG是通过计算和统计图像局部区域的梯度方向直方图来构成特征的。在特征提取过程中将目标图像看成一个窗口（Window），又将一个窗口分成若干个块（Block），而每一个块包含4个细胞单元（Cell）。对每一个细胞单元中的像素点，计算得到梯度的模和角度，并对此细胞单元进行直方图统计，得到一个9维向量。把一个块的特征向量联起来得到36维的特征向量，用块对窗口进行扫描，扫描步长为一个细胞单元。最后将所有块的特征串联起来，就得到了目标的特征。HOG constitutes features by calculating and counting the gradient direction histogram of the local area of the image. In the process of feature extraction, the target image is regarded as a window (Window), and a window is divided into several blocks (Block), and each block contains 4 cell units (Cell). For the pixel points in each cell unit, calculate the modulus and angle of the gradient, and perform histogram statistics on this cell unit to obtain a 9-dimensional vector. Connect the eigenvectors of a block to obtain a 36-dimensional eigenvector, and use the block to scan the window, and the scanning step is one cell unit. Finally, the features of all blocks are concatenated to obtain the features of the target.
（2）SVM训练(2) SVM training
SVM训练的过程主要包括准备训练样本集（包括证样本和负样本）、提取所有正负样本的HOG特征、为正负样本赋予标签（正样本标记为1，负样本标记为0）、用线性SVM进行训练得到分类器并将其保存为文本文件。由于要在各个方向对车辆进行识别，所以训练了两个分类器，分别用于识别车辆的正背面及侧面。The process of SVM training mainly includes preparing the training sample set (including proof samples and negative samples), extracting HOG features of all positive and negative samples, assigning labels to positive and negative samples (positive samples are marked as 1, and negative samples are marked as 0), and linear SVM is trained to get a classifier and save it as a text file. Since the vehicle needs to be recognized in all directions, two classifiers are trained to recognize the front, back and side of the vehicle respectively.
（3）SVM检测(3) SVM detection
为了让飞行器在与车辆的一定距离内都能检测到车辆，采用了多尺度的检测方法。首先对原始输入图片进行HOG特征提取，并用（2）中训练好的分类器进行车辆检测。若未能检测到车辆，则对输入图片进行一定尺度的缩放再次检测，直到检测到车辆或者输入图片大小比窗口小为止。在检测过程中，首先用检测车辆正背面的分类器进行检测，若识别到车辆，则可以判断飞行器可以拍摄到车牌，不需进行车辆方位检测。若没有识别到车辆，则继续使用检测车辆侧面的分类器进行检测，若识别到车辆的侧面，则将识别到车辆的矩形窗口作为感兴趣区域，进行车辆方位检测。In order to allow the aircraft to detect the vehicle within a certain distance from the vehicle, a multi-scale detection method is adopted. First, HOG feature extraction is performed on the original input image, and the classifier trained in (2) is used for vehicle detection. If the vehicle cannot be detected, the input image is scaled to a certain scale and detected again until the vehicle is detected or the size of the input image is smaller than the window. In the detection process, the classifier that detects the front and back of the vehicle is first used for detection. If the vehicle is recognized, it can be judged that the aircraft can capture the license plate, and there is no need for vehicle orientation detection. If the vehicle is not recognized, continue to use the classifier that detects the side of the vehicle for detection. If the side of the vehicle is recognized, the rectangular window that recognizes the vehicle is used as the region of interest for vehicle orientation detection.
步骤2、对识别到的车辆，检测其车轮轮廓形状Step 2. For the recognized vehicle, detect its wheel profile shape
车轮轮廓检测主要过程为图片边缘检测，椭圆轮廓拟合，车轮轮廓筛选几个步骤。The main process of wheel contour detection is image edge detection, ellipse contour fitting, and wheel contour screening.
（1）图片边缘检测(1) Image edge detection
只有灰度图片才可以进行边缘提取，但是直接度灰度图片进行边缘提取由于光照，阴影等的影响，可能会使车轮轮廓残缺或丢失，所以为了更好的保留和突出车轮的边缘，针对车胎为黑色的特点在HSV色彩模型下对感兴趣区域进行了二值化。通过在HSV色彩模型下对H通道进行检测，求得图片H通道的值为h，将h≤60区域的灰度置为0，其他区域置为255。通过该方法得到的图片能够将轮胎灰度置为0，轮毂灰度置为255，有效突出轮毂轮廓。Only grayscale images can be used for edge extraction, but the edge extraction of direct grayscale images may cause the wheel outline to be incomplete or lost due to the influence of light, shadow, etc., so in order to better preserve and highlight the edge of the wheel, for the tire The region of interest is binarized under the HSV color model for black features. By detecting the H channel under the HSV color model, the value of the H channel of the picture is obtained to be h, and the gray level of the h≤60 area is set to 0, and other areas are set to 255. The image obtained by this method can set the grayscale of the tire to 0 and the grayscale of the wheel to 255, effectively highlighting the outline of the wheel.
（2）椭圆轮廓拟合(2) Ellipse contour fitting
通过最小二乘拟合方法将检测到的边缘拟合成椭圆，但是由于图片中会检测到很多边缘，所以拟合到的椭圆不只有车轮的轮廓，要经过进一步筛选。The detected edge is fitted into an ellipse by the least squares fitting method, but since many edges are detected in the picture, the fitted ellipse is not only the outline of the wheel, but needs to be further screened.
（3）车轮轮廓筛选(3) Wheel profile screening
主要通过面积，位置和颜色三个特征对拟合到的椭圆进行筛选，最终得到代表车轮轮廓的两个椭圆。首先是面积满足下式时才判别为车轮：The fitted ellipses are screened mainly through the three features of area, position and color, and finally two ellipses representing the outline of the wheel are obtained. First of all, it is judged as a wheel only when the area satisfies the following formula:
Si/255≤Sw≤Si/25Si /255≤Sw ≤Si /25
其中感兴趣区域面积为Si，椭圆轮廓的外接矩形面积为Sw。The area of the region of interest is Si , and the area of the circumscribed rectangle of the ellipse is Sw .
其次是位置，代表车轮的椭圆的中心坐标应该位于感兴趣区域的下端，建立高斯模型：The second is the position. The center coordinates of the ellipse representing the wheel should be located at the lower end of the region of interest, and a Gaussian model is established:
其中μ为感兴趣区域的归一化底边纵坐标，y为归一化椭圆中心的纵坐标，并令σ2=1。则当f(y)≥0.8时，拟合到的椭圆才有可能代表车轮的轮廓。Where μ is the normalized bottom ordinate of the region of interest, y is the normalized ellipse center ordinate, and σ2 =1. Then when f(y)≥0.8, the fitted ellipse may represent the outline of the wheel.
最后是颜色，对于经过权利1得到的二值图片，符合车轮特点的椭圆轮廓中心不变，尺度放大到1.2倍得到的新轮廓，其上的像素点灰度值应该为0；所以对其灰度值进行统计，若为0的像素个数大于总数的90%，则判定为车轮椭圆。The last is the color. For the binary image obtained through entitlement 1, the center of the ellipse contour that conforms to the characteristics of the wheel remains unchanged, and the new contour obtained by enlarging the scale to 1.2 times, the gray value of the pixel on it should be 0; so for the gray If the number of pixels that are 0 is greater than 90% of the total, it is determined to be a wheel ellipse.
3、以步骤（2）的结果为根据，判断车辆与飞行器的相对方位，为飞行器自主对准车牌提供控制信息3. Based on the result of step (2), judge the relative orientation of the vehicle and the aircraft, and provide control information for the aircraft to autonomously align the license plate
通过计算车辆两个车轮中心的连线与水平的夹角来判断车辆与无人机之间的相对方位；当检测到两个代表车轮的椭圆边缘时，首先比较他们的大小，以较大者的中心为顶点，求得两椭圆中心连线与水平线的夹角θ0，则无人机需要较大车轮的方向旋转角度θ则可对准车牌，其中θ=90°-θ0。Determine the relative orientation between the vehicle and the UAV by calculating the angle between the line connecting the centers of the two wheels of the vehicle and the horizontal; when two ellipse edges representing the wheels are detected, first compare their sizes and choose the larger one The center of the ellipse is the vertex, and the angle θ0 between the line connecting the centers of the two ellipses and the horizontal line is obtained, then the UAV needs a larger wheel rotation angle θ to align with the license plate, where θ=90°-θ0 .
本发明的主要特点在于提出了一种通过检测车轮轮毂的轮廓并根据检测到的轮廓的大小和相对位置得到车辆方位角的方式。解决了此前大部分车辆方位识别算法只能识别到车辆的几个固定方位，不能得到连续的方位角的问题。该问题的解决为四轴飞行器准确判别与车辆的相对位置最终对准车牌提供了实现的可能。The main feature of the present invention is to propose a way to obtain the vehicle azimuth angle by detecting the outline of the wheel hub and according to the size and relative position of the detected outline. It solves the problem that most of the previous vehicle orientation recognition algorithms can only recognize several fixed orientations of the vehicle and cannot obtain continuous azimuth angles. The solution to this problem provides the possibility for the quadcopter to accurately determine the relative position with the vehicle and finally align it with the license plate.
附图说明Description of drawings
图1为经过HSV彩色模型二值化后的图片。Figure 1 is a picture after binarization by the HSV color model.
图2为车轮和车辆方位检测结果。Figure 2 shows the detection results of the wheel and vehicle orientation.
具体实施方式Detailed ways
下面结合附图和实施例，对本发明做进一步说明。The present invention will be further described below in conjunction with the accompanying drawings and embodiments.
本发明提出的四轴飞行器判别车辆方位的方法主要包括车辆识别，车轮识别和方向角计算，使用四轴飞行器在与车辆不同的相对位置拍摄视频并保存下来进行离线测试，视频大小为640*480。下面分别介绍具体实施方式：The method for judging the orientation of a vehicle by a four-axis aircraft proposed by the present invention mainly includes vehicle identification, wheel identification and direction angle calculation. The four-axis aircraft is used to shoot a video at a relative position different from the vehicle and save it for offline testing. The video size is 640*480 . The specific implementation methods are introduced respectively below:
1、车辆识别1. Vehicle identification
车辆识别主要通过基于HOG特征的SVM训练完成。用于识别车辆正背面的训练样本大小为96*112，用于识别车辆侧面的训练样本大小为112*64。正样本选取各个角度车辆的图片，负样本为各种不包含车辆的图片，例如建筑物，树木，自行车，行人等。样本总数为正样本：13736张，负样本：8649张。Vehicle recognition is mainly done through SVM training based on HOG features. The training sample size for identifying the front and back of the vehicle is 96*112, and the training sample size for identifying the side of the vehicle is 112*64. The positive samples select pictures of vehicles from various angles, and the negative samples are various pictures that do not contain vehicles, such as buildings, trees, bicycles, pedestrians, etc. The total number of samples is positive samples: 13736, negative samples: 8649.
SVM训练过程中，HOG特征提取的窗口大小即为目标大小也就是训练样本大小，细胞单元的大小为8*8，块的大小为16*16。在检测过程中，检测窗口以细胞单元大小为步长在待检测图片上滑动遍历图像，直到检测到车辆或者遍历完成。During the SVM training process, the window size of HOG feature extraction is the target size, which is the training sample size, the cell unit size is 8*8, and the block size is 16*16. During the detection process, the detection window slides and traverses the image on the image to be detected with the cell unit size as the step size, until the vehicle is detected or the traversal is completed.
车辆识别的结果为没有车辆、车辆的正背面、车辆侧面三种。若识别到车辆，则将识别到车辆的窗口作为感兴趣区域用于车轮检测。There are three types of vehicle recognition results: no vehicle, the front and back of the vehicle, and the side of the vehicle. If a vehicle is identified, the window where the vehicle is identified is used as the region of interest for wheel detection.
2、车辆方位检测2. Vehicle orientation detection
我们通过计算车辆两个车轮中心的连线与水平的夹角来判断车辆与无人机之间的相对方位。主要过程为图片边缘检测，椭圆轮廓拟合，车轮轮廓筛选和方位角计算几个步骤。能够检测的角度范围为90°~50°，当角度为50°时，可以满足拍摄到车牌的要求。We judge the relative orientation between the vehicle and the UAV by calculating the angle between the line connecting the centers of the two wheels of the vehicle and the horizontal. The main process is image edge detection, ellipse contour fitting, wheel contour screening and azimuth calculation. The angle range that can be detected is 90°~50°. When the angle is 50°, it can meet the requirements of capturing the license plate.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510562314.5ACN105184301B (en) | 2015-09-07 | 2015-09-07 | A kind of method that vehicle heading is differentiated using four-axle aircraft |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510562314.5ACN105184301B (en) | 2015-09-07 | 2015-09-07 | A kind of method that vehicle heading is differentiated using four-axle aircraft |
| Publication Number | Publication Date |
|---|---|
| CN105184301Atrue CN105184301A (en) | 2015-12-23 |
| CN105184301B CN105184301B (en) | 2018-06-26 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510562314.5AActiveCN105184301B (en) | 2015-09-07 | 2015-09-07 | A kind of method that vehicle heading is differentiated using four-axle aircraft |
| Country | Link |
|---|---|
| CN (1) | CN105184301B (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108227738A (en)* | 2017-12-28 | 2018-06-29 | 湖北电鹰科技有限公司 | A kind of unmanned plane barrier-avoiding method and system |
| CN108647670A (en)* | 2018-05-22 | 2018-10-12 | 哈尔滨理工大学 | A kind of characteristic recognition method of the lateral vehicle image based on support vector machines |
| CN108765444A (en)* | 2018-05-25 | 2018-11-06 | 哈尔滨工业大学 | Ground T shape Moving objects detection and location methods based on monocular vision |
| CN111626080A (en)* | 2019-02-27 | 2020-09-04 | 初速度（苏州）科技有限公司 | Vehicle detection method and vehicle-mounted terminal |
| CN114265499A (en)* | 2021-12-17 | 2022-04-01 | 交控科技股份有限公司 | Interaction method and system applied to customer service terminal |
| WO2022246767A1 (en)* | 2021-05-27 | 2022-12-01 | 华为技术有限公司 | Method and device for determining steering intention of target vehicle |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080071431A1 (en)* | 2006-09-19 | 2008-03-20 | Dockter Gregory E | Precision Approach Control |
| CN101547344A (en)* | 2009-04-24 | 2009-09-30 | 清华大学深圳研究生院 | Video monitoring device and tracking and recording method based on linkage camera |
| CN103985230A (en)* | 2014-05-14 | 2014-08-13 | 深圳市大疆创新科技有限公司 | Notification method, device and system based on image |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080071431A1 (en)* | 2006-09-19 | 2008-03-20 | Dockter Gregory E | Precision Approach Control |
| CN101547344A (en)* | 2009-04-24 | 2009-09-30 | 清华大学深圳研究生院 | Video monitoring device and tracking and recording method based on linkage camera |
| CN103985230A (en)* | 2014-05-14 | 2014-08-13 | 深圳市大疆创新科技有限公司 | Notification method, device and system based on image |
| Title |
|---|
| YUN HOU 等: "Autonomous target localization using quadrotor", 《CONTROL AND DECISION CONFERENCE （2014 CCDC），THE 26TH CHINESE》* |
| 张鹏 等: "用于违章停车车辆自动取证的四轴飞行器系统", 《太赫兹科学与电子信息学报》* |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108227738A (en)* | 2017-12-28 | 2018-06-29 | 湖北电鹰科技有限公司 | A kind of unmanned plane barrier-avoiding method and system |
| CN108227738B (en)* | 2017-12-28 | 2019-07-19 | 湖北电鹰科技有限公司 | A method and system for obstacle avoidance of unmanned aerial vehicle |
| CN108647670A (en)* | 2018-05-22 | 2018-10-12 | 哈尔滨理工大学 | A kind of characteristic recognition method of the lateral vehicle image based on support vector machines |
| CN108765444A (en)* | 2018-05-25 | 2018-11-06 | 哈尔滨工业大学 | Ground T shape Moving objects detection and location methods based on monocular vision |
| CN111626080A (en)* | 2019-02-27 | 2020-09-04 | 初速度（苏州）科技有限公司 | Vehicle detection method and vehicle-mounted terminal |
| CN111626080B (en)* | 2019-02-27 | 2022-06-24 | 魔门塔(苏州)科技有限公司 | Vehicle detection method and vehicle-mounted terminal |
| WO2022246767A1 (en)* | 2021-05-27 | 2022-12-01 | 华为技术有限公司 | Method and device for determining steering intention of target vehicle |
| CN114265499A (en)* | 2021-12-17 | 2022-04-01 | 交控科技股份有限公司 | Interaction method and system applied to customer service terminal |
| Publication number | Publication date |
|---|---|
| CN105184301B (en) | 2018-06-26 |
| Publication | Publication Date | Title |
|---|---|---|
| CN109993056B (en) | Method, server and storage medium for identifying vehicle illegal behaviors | |
| Yang et al. | Vehicle detection in intelligent transportation systems and its applications under varying environments: A review | |
| CN109740478B (en) | Vehicle detection and recognition method, device, computer equipment and readable storage medium | |
| CN101339601B (en) | License plate Chinese character recognition method based on SIFT algorithm | |
| CN105184301B (en) | A kind of method that vehicle heading is differentiated using four-axle aircraft | |
| CN109190444B (en) | An implementation method of a video-based toll lane vehicle feature recognition system | |
| CN103530600B (en) | Licence plate recognition method under complex illumination and system | |
| CN102968646B (en) | A license plate detection method based on machine learning | |
| CN102915433B (en) | Character combination-based license plate positioning and identifying method | |
| CN103093249B (en) | A kind of taxi identification method based on HD video and system | |
| Grbić et al. | Automatic vision-based parking slot detection and occupancy classification | |
| CN110619750A (en) | Intelligent aerial photography identification method and system for illegal parking vehicle | |
| CN105335702B (en) | A kind of bayonet model recognizing method based on statistical learning | |
| CN103136528B (en) | A kind of licence plate recognition method based on dual edge detection | |
| Prates et al. | Brazilian license plate detection using histogram of oriented gradients and sliding windows | |
| CN111078946A (en) | Bayonet vehicle retrieval method and system based on multi-target regional characteristic aggregation | |
| Sugiharto et al. | Traffic sign detection based on HOG and PHOG using binary SVM and k-NN | |
| CN107918775B (en) | Zebra crossing detection method and system for assisting safe driving of vehicle | |
| Singh et al. | Smart traffic monitoring through real-time moving vehicle detection using deep learning via aerial images for consumer application | |
| Wali et al. | Shape matching and color segmentation based traffic sign detection system | |
| Yao et al. | Coupled multivehicle detection and classification with prior objectness measure | |
| CN104143077A (en) | Image-based Pedestrian Target Retrieval Method and System | |
| CN102142090B (en) | Vehicle detection method and system | |
| CN112115800A (en) | Vehicle combination recognition system and method based on deep learning target detection | |
| Emami et al. | Real time vehicle make and model recognition based on hierarchical classification |
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |