Target Detection Method In Radar Traffic Scene Based On Deep Learning

Radar sensing technology is applied to urban road scenes to provide all-weather road sensing information for urban intelligent transportation,automatic driving,vehicle road coordination and other scenes.Target detection technology is an important basic ability of radar sensing technology,and its research value is self-evident.In recent years,with the continuous development of urbanization,the complexity and diversity of urban road scenes are increasing.In the face of complex urban traffic scenes,the traditional radar target detection methods have some disadvantages,such as low precision,slow speed and poor flexibility,which are difficult to meet the needs of practical application.Therefore,it is urgent to develop radar robust and real-time target detection technology suitable for urban traffic scenes.In order to improve the robustness and real-time performance of radar traffic scene target detection,this thesis mainly studies radar traffic scene target detection methods,including CFAR post clustering extended target detection algorithm,region based deep learning target detection algorithm and direct regression based deep learning target detection algorithm.The specific research contents are as follows:1)Aiming at the problem of radar target detection perceived by base stations in traffic scenes,this thesis introduces the method and process of pulse radar echo signal processing,expounds the principle of CFAR detection algorithm based on R-D data plane,and proposes a CFAR based post clustering target detection algorithm to solve the problem of extended target detection in traffic scenes;The detection performance of the proposed algorithm is verified by the measured data set,and the problems in the detection accuracy of the algorithm are analyzed.2)Aiming at the problems that the radar traffic scene R-D image has less target feature information and the targets are mostly variable scale weak targets,an improved regional proposal deep learning target detection algorithm model based on attention mechanism and pyramid feature fusion is proposed,which improves the accuracy of radar traffic scene target detection,and the detection accuracy of the proposed improved method is verified by the measured data.3)Aiming at the problem of real-time target detection in radar traffic scene,a model optimization method of YOLOv3 algorithm based on attention mechanism and uncoupled head anchor free detection model is proposed to realize real-time and robust target detection in radar traffic scene,and the improvement of detection accuracy and speed of the improved algorithm is verified by measured data.