Message passing based data association algorithm for multiple extended object Tracking

Abstract

Due to the development of sensor technology in recent years, it has become increasingly common for objects to occupy multiple units of sensor resolution. For tracking an extended object (EO) that produces an unknown number of measurements per time, the main challenge is to identify the origin of the measurements. Rather than enumerating all measurements partitions or association hypotheses, we adopt a message passing inference method to reduce the computational complexity of obtaining the marginal association probabilities. An overcomplete description of data association uncertainty has been introduced to obtain the marginal association probabilities with linear complexity in the number of measurements and targets. Based on factor graph theory and the sum-product algorithm (SPA), the proposed algorithm has a satisfied performance compared to other data association methods.