A variational bayes based extened object tracking method for 4D radar
Haowen Zhou, Qi Wang, Xiaoyue Huang, Majun Song, Ping Li, Jialong Jin, Sha Li, Jingzheng Li · IET conference proceedings. · 2024
Compared with traditional low-resolution 3D automotive radar, emerging 4D automotive radar has the significantly enhanced environmental perception capability and can provide richer point clouds, making it possible to estimate the object extension accurately. However, the radar measurements in polar coordinates exhibit a nonlinear relationship with the object motion and extension state modelled in the Cartesian coordinate system. To deal with the nonlinearities in the measurement equation, a novel extended object tracking algorithm using radar measurements is proposed. With the utilization of variational Bayesian inference, the approximate analytical estimation of the motion and extension state densities for the assumed elliptical extended object is obtained recursively. Furthermore, combined with the global nearest neighbour algorithm, the proposed method is generalized to the multiple object tracking scenarios, where the position gate adjusts its size according to the object extension estimation. By doing so, the occurrence of false associations can be reduced effectively when the tracked objects are closely spaced. Finally, the effectiveness of our approach is validated using the real test data of 4D radar.