Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The detection and rectification for identity-switch based on unfalsified control

Junchao Huang, Xiaoqi He Yebo Wu, Sheng Zhao

TL;DR

Unfalsified-control-based tracking (unfctrack) addresses identity-switch in MOT by detecting switches through a history-based metric $T_{spec}$ and rectifying them using pre-switch appearance features, while AMI mitigates ambiguous matching. The approach yields a modular framework that can augment existing trackers with IDSD, IDSR, and AMI. Evaluations on MOT17 and MOT20 under private-detection settings demonstrate partial success in detecting and correcting ID-switches, with AMI providing additional gains. This work introduces a practical pathway to maintain identity-consistent trajectories in challenging scenes and highlights the importance of leveraging historical appearance data for robust MOT.

Abstract

The purpose of multi-object tracking (MOT) is to continuously track and identify objects detected in videos. Currently, most methods for multi-object tracking model the motion information and combine it with appearance information to determine and track objects. In this paper, unfalsified control is employed to address the ID-switch problem in multi-object tracking. We establish sequences of appearance information variations for the trajectories during the tracking process and design a detection and rectification module specifically for ID-switch detection and recovery. We also propose a simple and effective strategy to address the issue of ambiguous matching of appearance information during the data association process. Experimental results on publicly available MOT datasets demonstrate that the tracker exhibits excellent effectiveness and robustness in handling tracking errors caused by occlusions and rapid movements.

The detection and rectification for identity-switch based on unfalsified control

TL;DR

Unfalsified-control-based tracking (unfctrack) addresses identity-switch in MOT by detecting switches through a history-based metric and rectifying them using pre-switch appearance features, while AMI mitigates ambiguous matching. The approach yields a modular framework that can augment existing trackers with IDSD, IDSR, and AMI. Evaluations on MOT17 and MOT20 under private-detection settings demonstrate partial success in detecting and correcting ID-switches, with AMI providing additional gains. This work introduces a practical pathway to maintain identity-consistent trajectories in challenging scenes and highlights the importance of leveraging historical appearance data for robust MOT.

Abstract

The purpose of multi-object tracking (MOT) is to continuously track and identify objects detected in videos. Currently, most methods for multi-object tracking model the motion information and combine it with appearance information to determine and track objects. In this paper, unfalsified control is employed to address the ID-switch problem in multi-object tracking. We establish sequences of appearance information variations for the trajectories during the tracking process and design a detection and rectification module specifically for ID-switch detection and recovery. We also propose a simple and effective strategy to address the issue of ambiguous matching of appearance information during the data association process. Experimental results on publicly available MOT datasets demonstrate that the tracker exhibits excellent effectiveness and robustness in handling tracking errors caused by occlusions and rapid movements.
Paper Structure (14 sections, 4 equations, 5 figures, 3 tables)

This paper contains 14 sections, 4 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related works
  3. Motion-based tracking
  4. Appearance-based tracking
  5. Other types of learning-based tracking
  6. Methods
  7. ID-switch detection (IDSD)
  8. ID-switch rectification (IDSR)
  9. Ambiguous match improvement (AMI)
  10. Experiments
  11. Experimental settings
  12. Testing and experiments results
  13. Limitations
  14. CONCLUSIONS

Figures (5)

  • Figure 1: Overview of the Unfctrack tracker's simplified channel diagram.
  • Figure 2: The key to identifying identity switch lies in establishing a time series of appearance information, which is used for subsequent IDSD and IDSR modules.
  • Figure 3: AMI module addressing the ambiguous match issue between trajectories and detection boxes. Firstly, the module discards low-confidence matches with a threshold greater than $d_\theta$=0.2, setting them to 1. For each row and column, the module also discards low-weight ambiguous matches. For example, Track3-Det1 and Track2-Det1 are compared, and since 0.18 is significantly greater than 0.03, the module discards the match with a weight of 0.18 and sets it to 1.
  • Figure 4: Experimental results of IDSD and IDSR on the MOT17-01 dataset. The two subfigures illustrate the increasing cosine cost and $T_{spec}$ after an ID-switch occurrence, as well as the variation of $T_{spec}$ after ID rectification. (a)The change of cosine cost.(b)The change of $T_{spec}$.
  • Figure 5: The three subfigures show the experimental image results at frames 24, 52, and 54.The IDSD model determines the ID-switch at frame 53, and the IDSD model rectify the ID at frame 54.