Multiple object tracking by incorporating a particle filter into the min-cost flow model

Abstract

A novel network flow model is proposed for multiple object tracking. Based on tracklets, only a short and reliable detection sequence is needed for an effective tracking. Our model fuses the local and global data association strategies to compensate for their respective shortcomings, which can be divided into two stages: A local stage and a global stage. In the local stage, we follow the tracking-by-detection framework to generate confident tracklets by employing a boosted particle filter. In the global stage, the data association problem is formulated as a Maximum-a-Posteriori (MAP) problem and solved by a typical min-cost flow algorithm. A double-step optimization is designed to solve the long term occlusion. The experimental results show that our method outperforms several state-of-the-art methods for multiple object tracking.