A New Approach for Train Driver Tracking in Real Time

Abstract

Find the main driver in the train cab and keep tracking him is the primary step for driver activity analysis in real time. But long-term robust human tracking is still a challenging task due to many practical factors. Most template-based tracking algorithms of templates often fail when encountering occlusion, pose variation and motion blur. In this paper, we propose a real-time algorithm based on correlation filters for long-term human tracking. To restore the false model caused by occlusion, we construct a conditional target re-detection mechanism. We adopted a high-confidence model update strategy to ensure the correct detection and avoid model drift problem. The experimental results indicate that the proposed algorithm is robust and accurate to variations in pose and heavy occlusion while runs at speed in real-time.