Vehicle Detection,Tracking And Distance Estimation For Autonomous Driving

In this paper,the problems of vehicle detection algorithm,distance estimation and image fusion of millimeter wave radar information based on monocular vision are studied.These problems include: 1)Using fixed confidence threshold to select vehicles leads to false alarm or false negative;2)road fluctuations leads to inaccurate monocular visual distance estimation;3)information fusion problem caused by inconsistent processing cycle and delay between visual algorithm and millimeter wave radar algorithm.To solve these problems,we study the adaptive confidence threshold algorithm for vehicle detection in deep learning,the monocular vision distance estimation algorithm by fusing vehicle pose information,and information matching between images and millimeter wave radar.The main contributions of this paper are:1.Analyze the influence of using fixed confidence threshold to choose vehicles in deep learning vehicle detection algorithm,including false alarm and false negative.Meanwhile,an adaptive confidence threshold algorithm is proposed to solve the problem of fixed confidence threshold,and we constructed data sets to test the performance.2.In the monocular vision distance estimation,the pose between the camera and the world coordinate system changes because of the road fluctuation.Therefore,the vehicle pose information is introduced to correct the conversion relationship between the world coordinate system and the camera,and a distance estimation algorithm based on vehicle pose information is proposed.To enhance accuracy of distance estimation for distant vehicles,we compute the width of the vehicle by data fitting and estimate the distance between vehicles.The experimental results show that our algorithms have a better performance.3.To solve the problem of information matching between vehicle detection results and millimeter-wave radar caused by the complexity of time,this paper designs a framework that applies vehicles tracking and combines millimeter-wave radar information.In this framework,millimeter-wave radar information is used to optimize the fast target tracking algorithm,and fast vehicles tracking method is used to match images and millimeter-wave radar information.