Multi-Bernoulli based track-before-detect with road constraints

Abstract

The random set based multi-Bernoulli filter is applied to a challenging low signal to noise track before detect scenario. Specifically we use the variant of the multi-Bernoulli filter that processes raw image observations. We add an additional layer of track management logic to output trajectories rather than point estimates. The tracker also exploits additional road map information by integrating the roads into the filtering likelihood. We show that this approach of using the image observation MeMBer filter with track management and road constrained model can yield an effective tracker for track before detect scenarios.