StrongSORT: Make DeepSORT Great Again

Abstract

Existing Multi-Object Tracking (MOT) methods can be roughly classified as tracking-by-detection and joint-detection-association paradigms. Although the latter has elicited more attention and demonstrates comparable performance relative to the former, we claim that the trackingby-detection paradigm is still the optimal solution in terms of tracking accuracy. In this paper, we revisit the classic tracker DeepSORT and upgrade it from various aspects, i.e., detection, embedding and association. The resulting tracker, called StrongSORT, sets new HOTA and IDF1 records on MOT17 and MOT20. We also present two lightweight and plug-and-play algorithms to further refine the tracking results. Firstly, an appearance-free link model (AFLink) is proposed to associate short tracklets into complete trajectories. To the best of our knowledge, this is the first global link model without appearance information. Secondly, we propose Gaussian-smoothed interpolation (GSI) to compensate for missing detections. Instead of ignoring motion information like linear interpolation, GSI is based on the Gaussian process regression algorithm and can achieve more accurate localizations. Moreover, AFLink and GSI can be plugged into various trackers with a negligible extra computational cost (591.9 and 140.9 Hz, respectively, on MOT17). By integrating StrongSORT with the two algorithms, the final tracker Strong-SORT++ ranks first on MOT17 and MOT20 in terms of HOTA and IDF1 metrics and surpasses the second-place one by 1.3 -2.2. Code will be released soon.

BoT-SORT: Robust Associations Multi-Pedestrian Tracking

“…More powerful detectors lead to the higher tracking performance and reduced the need for complex trackers. Thus, tracking-by-detection trackers mainly focus on improving data association, while exploiting deep learning trends [58,12].…”

“…Recently, the famous Kalman filter [8] with constant-velocity model assumption tends to be the popular choice for modeling the object motion [3,48,59,58,18]. Many studies use more advanced variants of the KF, for example, NSA-Kalman filter [13,12], which merge the detection score into the KF.…”

“…Discriminating and re-identifying (ReID) objects by deep-appearance cues [43,61,28] has also become popular, but falls short in many cases, especially when scenes are crowded, due to partially occlusions of persons. Separated appearance-based trackers crop the frame detection boxes and extract deep appearance features using the addition deep neural network [48,13,12]. They enjoy advanced training techniques but demand high inference computational costs.…”

“…s is the bounding box scale (area) and a is the bounding box aspect ratio. In more recent trackers [48,46,59,58,12] the state vector was changes to a eight-tuple, x = [x c , y c , a, h, ẋc , ẏc , ȧ, ḣ] . However, we found trough Figure 2.…”

Lightweight Indoor Multi-Object Tracking in Overlapping FOV Multi-Camera Environments

“…ByteTrack achieves the highest performance, but we wanted to take advantage of FastMOT’s skip function, which enhances the frame processing speed. In addition, there are various MOT methods [ 10 , 33 , 34 , 35 , 36 , 37 , 38 , 39 , 40 , 41 , 42 , 43 , 44 , 45 ]. In the experimental stage, the accuracy of global ID matching is measured by changing the skip parameter value.…”