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

BoostTrack: boosting the similarity measure and detection confidence for improved multiple object tracking

“…We demonstrate the effectiveness of proposed additions on MOT17 [41] and MOT20 [14] datasets. It has become a standard practice to apply camera motion compensation (CMC) [1,4,16,17,37,54] and interpolation of fragmented tracks [1,17,67,69] to MOT. By integrating CMC and gradient boosting interpolation from [67], we achieve comparable results with state of the art methods, without using time costly visual features and running at the speed of 65.45 FPS on MOT17 and 32.79 FPS on MOT20, on a desktop with one NVIDIA GeForce RTX 3090 GPU and AMD Ryzen 9 5950X 16-Core CPU.…”

“…It has become a standard practice to apply camera motion compensation (CMC) [1,4,16,17,37,54] and interpolation of fragmented tracks [1,17,67,69] to MOT. By integrating CMC and gradient boosting interpolation from [67], we achieve comparable results with state of the art methods, without using time costly visual features and running at the speed of 65.45 FPS on MOT17 and 32.79 FPS on MOT20, on a desktop with one NVIDIA GeForce RTX 3090 GPU and AMD Ryzen 9 5950X 16-Core CPU. Furthermore, by adding visual embedding to our system, which we refer to as BoostTrack+, at the expanse of longer run-time (15.35 FPS on MOT17 and 3.05 FPS on MOT20), we outperform all standard benchmark solutions.…”

“…We use linear interpolation implementation from [69] and do not set a maximum interval for interpolation. Tracks with less than 25 detections are not interpolated (default value from [69]), and we further improve interpolation results by applying gradient boosting interpolation from [67]. As in [1,37,69], we use FastReID [25] as a visual embedding model, and apply Dynamic Appearance embedding update from [37].…”

“…As the standard practice, Boost-Track uses camera motion compensation (CMC) and interpolation to connect fragmented tracks. We use gradient boosting interpolation (GBI) from [67], but we also show results obtained using linear interpolation (LI). Finally, we add appearance similarity (AS) and show the effect of added components on MOT17 and MOT20 datasets in table 3.…”