Research On Fusion Strategy Based On Monocular Vision And Millimeter Wave Radar

Self-driving vehicles will encounter many safety problems such as missing targets and unresponsive responses during driving,and thus,it is necessary to perform correct detection and recognition of the surrounding environment in order to achieve the purpose of precise navigation.The detection and recognition of obstacles during navigation helps to determine the trajectory of the vehicle entering the safe driving area.Therefore,it is necessary to fuse data from different sensors to achieve accurate detection of the target.Millimeter-wave radar(MMWR)has the advantages of not being affected by extreme weather and strong penetration,but due to insensitivity to small targets,MMWR will have the phenomenon of near-track target loss.Monocular vision can provide abundant category information,However,it is seriously disturbed by the environment,and its visual field is severely limited.Therefore,this thesis proposes a fusion strategy based on monocular vision and MMWR,which can provide more information for MMWR by using monocular vision to detect and identify short-track targets.The fusion of monocular vision and MMWR can provide drivers with more timely warning of dangerous areas by detecting the lost radar target points.The research content of this thesis is mainly divided into the following three points.(1).This thesis presents a target detection scheme based on millimeter wave radar.According to the principle of MMWR target detection and signal processing,a track initiation algorithm,a clustering algorithm and a kalman filtering algorithm are proposed respectively,to complete the MMWR clutter noise filtering,vehicles and pedestrians data correlation and multiple target tracking.(2).This thesis proposes a monocular vision-based target recognition algorithm to compare the performance of the YOLOv3 algorithm with commonly used multi-label image classification algorithms,and to make our own label dataset based on the VOC dataset.In order to prevent the impact of light changes on visual classification,the MSR algorithm is used to enhance the input image.By changing the output structure of the input layer and the fully connected layer of the YOLOv3 algorithm,the monocular vision algorithm trains its own data set and then compares the performance with other classification algorithms to verify the accuracy of the algorithm in this thesis.(3).This thesis proposes a MMWR fusion system based on monocular vision.By analyzing the shortcomings of the traditional coordinate conversion calibration method,a new spatial calibration method using the principle of monocular visual distance measurement is innovatively proposed.Aiming at the problem that the sampling frequencies of the MMWR and the monocular vision are not synchronized,the fusion algorithm utlizes the time stamp as a reference and uses a multi-threaded approach to achieve synchronization of MMWR data and camera data in time.Finally,this thesis applies the fusion system to road test in the actual traffic environment.In addition,this thesis analyzes and verifies the matching target results of the fusion system and the warning results of the missed detection area and the analysis and verification results show that the success rate of the missed detection target of the fusion system is about 93%.By comparing the detection results of the single sensor with the detection results of the fusion system,the feasibility of the fusion system proposed in this thesis in the actual highway scenario is proved.