Coupled data association and l1 minimization for multiple object tracking under occlusion

We propose a novel multiple object tracking algorithm in a particle filter framework, where the input is a set of candidate regions obtained from Robust Principle Component Analysis (RPCA) in each frame, and the goals is to recover trajectories of objects over time. Our method adapts to the changing appearance of objects, due to occlusion, illumination changes and large pose variations, by incorporating a l1 minimization-based appearance model into the Maximize A Posterior (MAP) inference. Though L1 trackers have showed impressive tracking accuracy, they are computationally demanding for multiple object tracking. Conventional data association methods using simple nonparametric appearance model, such as histogram-based descriptor, may suffer from drastic changing object appearance. The robust tracking performance of our approach has been validated with a comprehensive evaluation involving several challenging sequences and state-of-the-art multiple object trackers.