Score refinement for confidence-based 3D multi-object tracking

Abstract

Multi-object tracking is a critical component in autonomous navigation, as it provides valuable information for decision-making. Many researchers tackled the 3D multiobject tracking task by filtering out the frame-by-frame 3D detections; however, their focus was mainly on finding useful features or proper matching metrics. Our work focuses on a neglected part of the tracking system: score refinement and tracklet termination. We show that manipulating the scores depending on time consistency while terminating the tracklets depending on the tracklet score improves tracking results. We do this by increasing the matched tracklets' score with score update functions and decreasing the unmatched tracklets' score. Compared to count-based methods, our method consistently produces better AMOTA and MOTA scores when utilizing various detectors and filtering algorithms on different datasets. The improvements in AMOTA score went up to 1.83 and 2.96 in MOTA. We also used our method as a late-fusion ensembling method, and it performed better than voting-based ensemble methods by a solid margin. It achieved an AMOTA score of 67.6 on nuScenes test evaluation, which is comparable to other state-of-the-art trackers. Code is publicly available at: https://github.com/cogsys-tuebingen/CBMOT.

Immortal Tracker: Tracklet Never Dies

“…Different from previous works [1,7,12,21,23,24,27,28] which terminate a tracklet after several frames since its last successful association, our Immortal Tracker always maintains tracklets for objects even if they are invisible. The maintained tracklets will repeatedly predict their locations in the future frames with their last estimated states.…”

“…Tracklets that lost their targets for serveral(typically less than 5) frames will be terminated. [1,21] proposed to initialize and terminate tracks depending on their confidence score estimated from the confidence of their associated detections. However, they will still permanently terminate the tracklets that fail to be associated with new detections.…”

“…Ideally, a tracklet should be maintained until its target leaves the scene. Previous works [1,7,12,21,23,24,27,28] Figure 1. Bird's Eye View(BEV) tracking visualization of vehicles with their predicted trajectories.…”

TripletTrack: 3D Object Tracking using Triplet Embeddings and LSTM

“…[26] presents a tracking pipeline that exploits 2D and 3D world space motion consistency to improve long-term tracking under occlusions. [2] proposes a confidence-based method for initialization and termination of tracklets. State-of-the-art performance can also be achieved without using the tracking-by-detection paradigm [55].…”

“…Surprisingly, our motion model is sufficient to outperform existing methods, providing an additional evidence of the internal LSTM representation effectiveness. 2 10 meters is the value that maximizes the metrics performance.…”