CSE445Computer Vision
Tutorials/Object Tracking and Multi-Object Tracking/Appearance, Occlusion, and Robust Identity Tracking
Previous tutorialDetection-to-Track Association: From SORT to ByteTrackNext tutorialTracking Evaluation and Intelligent Video Analytics
TR-03Intermediate–Advanced 3 Hours

Appearance, Occlusion, and Robust Identity Tracking

Extend motion-only tracking with appearance representations, re-identification, camera-motion compensation, and observation-centric motion modeling. Students compare DeepSORT, BoT-SORT, OC-SORT, and related robust tracking strategies.

PythonPyTorchOpenCVscikit-learnYOLO26 detections

Code Structure Note

These notes are provided only to help students understand the code structure. Students should review, optimize, and adapt the code before using it in assignments, projects, or experiments.

Topics Covered
  1. 1Appearance embedding and re-identification
  2. 2Cosine similarity
  3. 3Motion–appearance fusion
  4. 4Occlusion and identity switch
  5. 5Track fragmentation and lost-track recovery
  6. 6Camera motion and homography
  7. 7Global motion compensation
  8. 8Observation-centric motion estimation
  9. 9Duplicate-track suppression

Learning Objectives

  • Explain when motion-only association becomes ambiguous
  • Use appearance embeddings for re-identification and recovery
  • Compare DeepSORT, BoT-SORT, and OC-SORT conceptually
  • Analyze camera motion, occlusion, fragmentation, and duplicate identities

Key Takeaways

  • Appearance supports identity recovery but depends on domain-appropriate features
  • Camera compensation separates global image motion from object motion
  • Observation-centric updates help with abrupt and nonlinear motion
  • No tracker is universally best; scene and motion assumptions matter

Tutorial Infographic

Hands-On Exercises

  1. 1Visualize appearance embeddings before and after occlusion
  2. 2Estimate camera motion from background features
  3. 3Compare track recovery across multiple robust trackers
  4. 4Write a failure analysis for one identity switch

Tools & Libraries

PythonPyTorchOpenCVscikit-learnYOLO26 detections
Previous tutorialDetection-to-Track Association: From SORT to ByteTrackNext tutorialTracking Evaluation and Intelligent Video Analytics