Robust Vehicle Tracking Method Based on Radar and Vision Fusion
Aiming at the problem of missed detection and false detection in the actual working environment of a single sensor,a robust vehicle tracking method based on radar and vision fusion is proposed.Firstly,a target recognition network called R-Densenet based on multi-dimensional data which are outputted by millimeter-wave radar is constructed.To obtain the radar identification results,multi-dimensional data detected by millimeter-wave radar is input to R-Densenet neural network.Then,The image taken by camera is input into the YOLO-V4tiny network to get the camera recognition results.The original target points are fused by the fusion rule,this step is to filter out the radar false detection points and noise points.Then the successful fu-sion points are continuously tracked by using the real target survival judgment method.During the tracking process,false detec-tion and missed detection of millimeter-wave radar or camera are captured and the lost data are supplemented by Kalman filter,YOLO re-identification and other methods.The results of the experiment show that the fusion method can effectively eliminate the false target,and has good robustness for single sensor missed detection,single sensor false detection and two sensors missed detec-tion at the same time during tracking,so the tracking is relatively stable.
Vehicle TrackingMillimeter-Wave RadarVisionInformation FusionTime and Space Alignment
NETL
NSTL
万方数据
