Adaptive multi-cue based particle swarm optimization guided particle filter tracking in infrared videos

• A multi-cue weight self-adjusting strategy is explored.
• An effective convergence criterion for particle swarm optimization is proposed.
• The multi-cue weights are estimated by the random particles at the start frame.

This paper presents a multi-cue based particle swarm optimization (PSO) guided particle filter (PF) tracking framework. In the proposed tracking framework, PSO is incorporated into the probabilistic framework of PF as an optimization scheme for the propagation of particles, which can make particles move toward the high likelihood area to find the optimal position in the state transition stage, and simultaneously the newest observations are utilized to update the relocated particles in the update stage. Furthermore, likelihood measure functions employing multi-cue are explored to improve the robustness and accuracy of tracking. Here, each cue weight is self-adaptively adjusted by PSO algorithm throughout the tracking process. Experiments performed on several challenging public infrared video sequences demonstrate that our proposed tracking approach achieves considerable performances.