Research On Front Vehicle Detection System Based On Millimeter Wave Radar And Vision Fusion

In recent years,with the development of intelligent driving towards high-level automatic driving,the vehicle environment sensing system has gradually transitioned from single sensor detection to multi-sensor fusion detection.Based on this background,this paper studies a vehicle detection system based on millimeter wave radar and visual fusion.Firstly,the effective target acquisition method of millimeter wave radar is studied.The consistency and decision of effective target vehicles are realized by thirdorder classical Kalman filter and life cycle algorithm;Secondly,considering the balance between detection speed and accuracy,the lightweight yolov4 is selected as the basic framework.Considering the lack of small target detection ability of the YOLOv4-tiny,a small target detection module based on feature pyramid network is proposed.The experiment on edge device shows that the detection accuracy is improved by 2.16% on the premise of ensuring the detection speed;Then,A target detection framework based on fusion strategy is proposed,which realizes the spatial and temporal data fusion of radar and camera,and generates the region of interest of radar target in visual image.The radar and visual target level fusion is realized by the method based on intersection over union,and the decision output is carried out;Finally,taking NVIDIA jetsontx2 as the edge device platform,the software and hardware environment of the experiment is built,and the data under different working conditions are collected synchronously.Experiments show that the fusion strategy algorithm proposed in this paper can detect vehicles under different working conditions.Compared with a single millimeter wave radar and visual detection scheme,the accuracy is improved by 13.23% and 4.04%respectively.Compared with other millimeter wave radar and visual fusion schemes,the accuracy is improved by 1.06% and the missed detection rate is reduced by 1.57%.