Tracking and Fusion of Multiple Detections for Multi-target Multi-sensor Tracking Applications in Urban Traffic

Recently, high-resolution sensors capable of multiple detections (MDs) per object are available for perception applications in autonomous or semi-autonomous vehicles. Conventional multi-target tracking (MTT) approaches start with the point-target assumption and thus cannot be applied directly to the MDs of high-resolution sensors. A popular solution widely used in literature starts with a measurement partitioning approach, followed by repurposing conventional tracking algorithms to accommodate the resulting partitions. However, the computational requirement increases combinatorially, especially under multi-sensor applications that also independently return multiple radar reflections as in the automotive radar sensors used in this work. Thus, a hybrid approach that combines a clustering technique (such as DBSCAN) to alleviate the computational complexity and an MD tracking scheme that admits multiplicity of the target detections is employed. The resulting method is applied to simulated and practical target tracking scenarios. We also validate the performance of the proposed MD tracker configurations based on filtering (estimation) accuracy and speed of computation, both of which are of utmost importance for real-time applications.