Research On Intelligent Analysis System Of Traffic Video Based On Object Detection And Tracking

The intelligent analysis for traffic video using computer vision technology is a research hotspot in the ITS(Intelligent Transportation System).By detecting and tracking the vehicle object in traffic video,object location and trajectory information could be obtained.The above result is processed to generate real-time traffic parameters and traffic incident information.The goal of intelligent analysis for the traffic video would be realized.This paper takes highway traffic video as the research object,and adopts the deep learning-based object detection method and multi-object tracking technology to realize a effective intelligent analysis system for highway videos.The paper includes the following:Object detection: This paper selects YOLOv3 as the basic framework of object detection,which has high accuracy and real-time performance,a dataset including 11,238 sample images from highway traffic scene is annotated.Based on this dataset,transfer learning is used to train,test and optimize the model to make the network optimal.The experimental results show that the average detection accuracy of the object detection algorithm is 89.2%,and the detection speed is 29 FPS,which meets the practical application requirements.Multi-object tracking: Combing with the object detection results,a data association method based on IoU similarity measurement and row-column optimization principle is proposed.This method realizes the correlation evaluation between object and trajectory,and designs corresponding trajectory processing strategy for different association results,and continuous tracking of multiple vehicle objects is achieved.Through experimental verification,the tracking algorithm of this paper can effectively deal with the problems such as false detection,undetected fault and occlusion.Intelligent analysis system of traffic video: Based on the object detection and tracking,the intelligent analysis system for highway traffic video is realized.The system can determine the traffic flow direction and set the traffic flow section detection line automatically using the object trajectory information,and obtain traffic parameters more accurately and detect traffic incidents.The experimental has verified that the system has certain practical value.