GM-PHD Filter based Multi-target Tracking Method for Radar and Monocular Camera

Target tracking is one of the most important technologies in intelligent vehicle environment perception. Because of complicated environment and the unknown interference, there are many problems in practical application, such as, the single sensor tracking method cannot effectively meet the application requirements, so the millimeter wave radar and camera fusion method is an effective method. But considering that large targets may correspond to multiple radar echoes, existing data association methods can't effectively address this situation. An improvement Gaussian mixture probability hypothesis density (GM-PHD) multi-target tracking algorithm based on monocular camera and radar is proposed. The camera target is converted to polar coordinates and associated based on the angle. Associating the target of radar and camera in polar coordinate by angle range of camera target, and then divide the measurement set into several levels, and set the confidence for each level. Combining the radial range of the millimeter-wave radar with the azimuth of the camera to estimate the position of the target more accurately, then the fusion measurement data is introduced into the improved GM-PHD multi-target tracking algorithm to update the multi-target state. The real target data are used to verify the tracking algorithm. The experimental results show that the proposed tracking algorithm can effectively improve the tracking accuracy and robustness.