Research On Target Recognition Method In Front Of Boarding Bridge By Fusion Of Millimeter Wave Radar And Vision

With the advancement of science,technology and society,and the rapid development of unmanned driving technology,the boarding bridge is also developing in the direction of automatic docking.As the "eyes" of unmanned driving,environmental sensors generally include millimeter wave radar,ultrasonic radar,laser radar,and cameras,but a single sensor has limitations in its perception of the environment due to various reasons.Multi-sensor fusion technology can integrate the advantages of each sensor,realize mutual information complementation,and provide information redundancy for the system.Therefore,this paper proposes a target recognition method in front of the boarding bridge based on the mainstream sensors of the two perception systems.The research focus and innovation points of this paper are as follows:1.Target recognition of millimeter wave radar and machine vision.Starting from the principle of radar recognition,preliminary radar data screening and consistent decision-making are carried out for the target in front of the boarding bridge.In machine vision,YOLOv4 is used as the target recognition framework,and the model is trained for the boarding bridge application scenarios.2.Research on fusion recognition method.Establish radar coordinate system,world coordinate system,camera coordinate system and pixel coordinate system,realize the fusion of radar and visual space through joint calibration and coordinate system conversion method,perform temporal fusion through the frame extraction method,and finally calculate the image perception through the fusion model Region of interest(ROI).3.Research on ROI area removal algorithm based on union search.Analyzed the overlap type of the ROI generated by the radar on the image,and judged the overlap type based on the merge collection,and proposed a method for de-duplication of the overlap area of the radar target rectangular frame,which reduced the false detection rate,the re-detection rate and the ROI area to a certain extent.Area.4.Research on ROI area combination method based on two-dimensional rectangular box packing problem(2DRP).Compared with the traditional fusion method,the 2DRP heuristic algorithm is improved,and a ROI region image combination method based on the improved box algorithm is proposed,which removes the invalid target image region,and improves the efficiency and accuracy of fusion recognition to a certain extent.rate.From the perspective of practical application,this paper studies the radar and vision fusion target recognition algorithm in front of the boarding bridge,which has certain research and application significance in improving the performance of the environment perception system of the boarding bridge automatic driving.