Research On Vehicle Multi-target Detection And Tracking In Video

As an important part of the traffic supervision system,road video surveillance can not only monitor the passing vehicles within the monitoring range in real time,but also achieve real-time warning of sudden traffic accidents.It is also possible to process video frames through corresponding intelligent algorithms combined with knowledge of digital image processing,and to automatically detect,identify and track passing vehicles in real time to achieve real-time grasp of effective information such as traffic volume and license plate numbers within the range.Finally intelligently monitor the behavior of passing vehicles.In order to improve the accuracy and timeliness of vehicle detection,license plate recognition and tracking,this article simulates road monitoring and acquisition vehicle video,firstly performs equalization processing and grayscale processing on the collected video;and then identifies the road lane line through edge detection.With segmentation,the road surface on which the current camera located is extracted as a region of interest;then a hybrid detection model based on the entropy method for weighting Adaboost and frame difference method is proposed to detect and locate vehicles in the area;After acquiring the detected vehicle target,the detected vehicle is tracked using an improved KCF algorithm;the acquired vehicle image is then preprocessed,and the preprocessing includes binarization,erosion,expansion,and edge detection of the vehicle image.In this process,a mathematical morphological coarse positioning combined with pixel location and scanning precision positioning is used to locate the license plate.The Radon transform is used to perform tilt correction on the tilted license plate,and the license plate character is segmented using the projection method.The segmentation characters are normalized.Finally,based on the traditional template matching method,this paper proposes an improved template matching method to identify the license plate characters.The experimental results show that the detection model proposed in this paper can effectively filter out other non-target objects and effectively detect and screen motor vehicles.At the same time,compared with the traditional optical flow method and background modeling method at the same time under the same conditions,the experimental results show that the algorithm used in this paper is superior to the other two algorithms in detection speed and accuracy..The detection accuracy of the detection model used in this paper can reach percentage of 96.25.The improved KCF tracking method can effectively overcome the occlusion between vehicle targets,and can track multiple vehicles in a stable and efficient manner.The license plate positioning method can quickly and accurately locate the license plate of the acquired vehicle;the used character segmentation method can effectively divide the license plate characters and effectively solve the problem of character adhesion;theproposed improved character recognition method is compared with the traditional template matching algorithm.Can effectively overcome the misjudgment of similar characters,the license plate recognition accuracy can reach percentage of 95.5.