Design Of Target Recognition System Based On Fusion Of Millimeter-wave Radar And Machine Vision

The popularization of motor vehicles facilitates people’s travel and improves their living standards,but it also causes a large number of traffic accidents.The driving assistance system can effectively identify the target of the specific area of the vehicle,give early warning before the possible traffic accident,and reduce the probability of the accident.The environment perception technology plays an important role in the automobile driving assistance system,which is one of the key research fields in the automobile industry.In practical applications,a single sensor cannot detect the target in the scene stably,robustly,all-time and comprehensively because of its working principle and stability.Multi sensor fusion is an important research direction to solve this problem.In this paper,a target recognition system is designed based on millimeter wave radar and vision sensor,which can recognize the vehicles,pedestrians and other targets in the road ahead and obtain the location,speed and other information.The main research contents are as follows:According to the vehicle operation scenario,system development cycle and actual cost,select the appropriate hardware and software.In terms of hardware selection,the characteristics,performance and cost of each sensor are analyzed and compared to determine the type of sensor.Then,according to the sensor category and industry related information,the mature and stable sensor model is selected to improve the stability and reliability of the system.In terms of software configuration,based on the difficulty of algorithm development,the richness of data and the convenience of debugging,the ROS framework based on Linux system is selected,and C++ is used for algorithm development.The system can meet the target recognition requirements of different scenes,different times and different weather conditions,and the cost is also controlled at a reasonable level.Aiming at the problems of poor clustering effect of millimeter wave radar targets,this paper proposes an adaptive clustering algorithm based on DBSCAN algorithm.Firstly,the point cloud signal sent by millimeter wave radar is collected by analytic program,then the clustering radius of point cloud is adaptively adjusted according to k-nearest neighbor distance and radar reflection cross-section area,so that the algorithm can adjust the clustering parameters according to the point cloud sparsity and reflection intensity,and improve the clustering effect of target.Aiming at the problem of cluster target glint of radar and visual target jump,the tracking process of radar target and visual target is designed.In target tracking of two sensors,the nearest neighbor method is used for data association,and the interactive multi model algorithm is used for tracking filtering,but the tracking threshold and tracking parameters need to be set according to the characteristics of the sensor.In order to make full use of the complementarity between millimeter wave radar and vision sensor,a multi-sensor fusion method combining feature level and decision level is proposed.Track association determines the fusion mode:the feature level fusion is used for the correlated tracks,and the federated filter is used to fuse the radar tracks and the visual tracks;the decision level fusion is used for the non-correlated tracks,and the appropriate uncorrelated targets are selected as the fusion targets according to the radar targets confidence and the high confidence features of the visual targets.The test results show that the system can identify the front target accurately and stably,and basically meet the needs of assistant driving perception.