NCT:noise-control multi-object tracking

Abstract Multi-Object Tracking (MOT) is an important topic in computer vision. Recent MOT methods based on the anchor-free paradigm trade complicated hierarchical structures for tracking performance. However, existing ignore noise detection, data association, and trajectory reconnection stages, which results serious problems, such as missing detection of small objects, insufficient motion information, drifting. To solve these this paper proposes Noise-Control Tracker (NCT), focuses noise-control design reconnection. First, a prior depth denoise method introduced to suppress fusion feature redundant noise, can recover gradient information heatmap features. Then, Smoothing Gain Kalman filter designed, combines Gaussian function with adaptive observation coefficient matrix stabilize mutation gain. Finally, address drift issue, boosting context mechanism realizes effectively fill gaps trajectories. With assistance plug-and-play method, experimental MOTChallenge 16 &17 datasets indicate that NCT achieve better performance than other state-of-the-art trackers.