Research On Multi-target Detection And Tracking Algorithm Of Intelligent Vehicle In High-density Traffic Scenario

Intelligent vehicle in reducing the driving load,improve the active safety and improve traffic efficiency and lower energy consumption has great potential,not only became the strategic direction of the global auto industry development,also is the historical opportunity of automobile power construction in our country,is a new generation of information technology,high-end equipment manufacturing,new materials,new energy and other strategic emerging industries innovation integrated carrier.Environment awareness technology is an important basis for intelligent vehicles and the bottleneck for intelligent vehicles to achieve high-level autonomous driving.Detection and tracking technology can provide perceptual output for subsequent positioning and mapping,which is the basis for subsequent advanced perception tasks and plays an important role in improving the safety of intelligent vehicles.This dissertation mainly considers the fast matching of object detection and identification in high-density traffic scene,in the existing visual detection algorithm and multiple target tracking algorithm on the basis of the targeted improvement,in order to meet the smart car perception tasks as the goal,the performance of the required accuracy and calculation speed is a high-density traffic scene more research of target detection and tracking,concrete research content is as follows:(1)A YOLO＿Free detection algorithm for high-density traffic scenes is proposed to achieve multi-target detection in high-density traffic scenes.The prediction method without anchor frame of multi-branch structure is used to detect high-density target,which solves the problem of insufficient performance of high-density target detection.Double random occlusion filling method is used to solve the problem of insufficient generalization caused by sparse occlusion in highdensity traffic scenes.DIo U loss function is used to solve the problem of poor detection effect caused by too close distance of objects in high-density traffic scenes.Experimental results show that the proposed algorithm not only improves the detection accuracy by 5.1% in high-density traffic scenarios,but also has a detection speed like the original algorithm.(2)A multi-target tracking algorithm combined with YOLO＿Free detection algorithm is proposed to achieve real-time multi-target tracking in high-density traffic scenarios.An identity mapping method based on Transformer structure is used to solve the problem of target movement uncertainty.The historical appearance features are used to correct the current appearance to solve the problem of appearance loss caused by occlusion.Low confidence detection frame is used for secondary association to solve the problem of high number of identities switching caused by missed detection.Using Transformer linearization acceleration to solve the problem of slow model calculation speed.Experimental results show that the proposed algorithm can not only improve the high-order tracking accuracy by 21.3% in high-density traffic scenarios,but also meet the demand of real-time speed.(3)Perform performance tests based on jiangdu Intelligent Line 2 real vehicle test platform of the research group to verify the feasibility of the algorithm.The code is based on python3.8 and C++ implementation,the construction framework of deep learning is Pytorch1.8.0,the operating system of Ubuntu18.04 is used,the development and debugging tool is Pycharm,and Py Qt5 is used to visualize the algorithm.The actual vehicle test shows that the proposed method can effectively realize the tracking of pedestrians and vehicles in high-density traffic scenarios and has high engineering value.