Research On Vehicle Detection Technology Based On Millimeter Wave Radar And Video Fusion

With the dramatic increase in car ownership in China over the past decade,traditional traffic systems are facing great challenges and effective traffic monitoring is crucial to the traffic system.Traffic monitoring systems usually use a single radar or camera sensor for traffic monitoring,which cannot sense the vehicle information comprehensively and accurately.In order to solve these problems,this paper proposes a millimeter wave radar and camera information fusion technique to achieve effective monitoring of traffic vehicles.The main work consists of the following parts:(1)The acquisition of effective targets of millimeter wave radar and video vehicles: the selection and working principle of millimeter wave radar are elaborated,and the life-cycle algorithm is designed to process the radar data and obtain effective vehicle speed and relative distance information for the problem that millimeter wave radar is prone to empty targets and invalid targets.(2)To address the problems of large model and poor real-time performance of existing vehicle detection algorithms,vehicle detection is performed using the improved YOLOv5-based algorithm,which uses Shuffle Net V2 network to replace the backbone structure of YOLOv5,reduces the number of parameters and computation of the model,improves the detection accuracy by adding SE attention mechanism,and achieves the lightweight improvement of the model.Experiments show that the size of the improved model is 8.5M,which is 40% smaller than the original algorithm model,and the detection accuracy is improved by 1.1%.The visual detection provides the basis for subsequent fusion,ensures the vehicle detection for rapidity,and creates favorable conditions for future deployment in mobile embedded devices.(3)Vehicle detection based on millimeter wave radar and vision information fusion.Firstly,the conversion relationship between radar and camera coordinate system is studied for the problem that millimeter wave radar data and camera data are in different coordinates,and the spatial alignment of the coordinate system is realized by translation rotation transformation and other relationships,then the temporal alignment of millimeter wave radar and camera is carried out by using Lagrangian interpolation method,and finally the joint calibration of sensors is completed.The fusion method based on intersection and parallel ratio is used to correlate the radar detection target and visual detection target in the same frame,and on the basis of coordinate calibration,the vehicle target localization is integrated into the coordinate system of camera and radar to obtain the accurate position and speed of the vehicle.The field validation shows that the vehicle detection accuracy of the fusion algorithm is 95.5%,which is 14.3% and 4.2% better than the single radar and vision detection algorithm.