Interaction-Aware Labeled Multi-Bernoulli Filter with Road Constraints

Abstract

Vehicle tracking is vital in many applications related to vehicle automation and surveillance. However, the realistic information regarding external influences on a vehicle’s motion is often disregarded. Many applications assume the motion of vehicles to be independent of all external factors. However, vehicles often continuously interact with other close vehicles, especially with their front vehicle. The interaction-aware labeled multi-Bernoulli (IA-LMB) filter has been explicitly designed to modify the LMB filter to incorporate distance-based interactions among targets, which is the most commonly occurring form of vehicle interaction. In this paper, we propose a new method to incorporate road-related information within the target state in the IA-LMB filter to depict further the efficacy of including realistic constraints on target motion for multi-object tracking. For a carefully designed synthetic scenario with multiple vehicle interactions and location influence on the road, we have depicted the advantages of the proposed method. Performance comparison has been conducted regarding the optimal sub-pattern assignment (OSPA) metric for the LMB filter, IA-LMB filter and the proposed method. Results show that including road information within the filtering process further enhances tracking accuracy.