Continuous Trajectory Tracking Across Sensors at Intersection using Dynamic Target Set within the Overlapping Area

Across-sensor tracking is an important technology to realize global perception in the field of next-generation intelligent transportation. However, identifying targets within an overlapping area and merging trajectories from different sensors remain challenging. Previous methods require high labor costs to manually identify an overlapping area and neglect the problem of noise and occlusion when data present inconsistency. To address these issues, this paper proposes a novel across-sensor tracking method at intersection. The proposed method adaptively identifies the target set for the overlapping area based on the Gaussian product of target states from two sensors and fuses trajectories based on Arithmetic Average algorithm. Additionally, the paper improves the birth model of the Poisson Multi-Bernoulli Mixture filter in order to handle complex intersection situations. Extensive simulation and real-world experiments evaluate the method's effectiveness, and its superior performance compared to previous methods.