Tracklet Association by Online Target-Specific Metric Learning and Coherent Dynamics Estimation

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Multi-person tracking is crucial for analyzing complex scenes.
• Existing methods struggle with occlusions and identity switches.
• Long-term tracking requires robust association of track fragments.

Purpose of the Study:

• To develop a novel method for tracklet association in long-term multi-person tracking.
• To improve tracking accuracy by preventing identity switches and recovering missed detections.
• To effectively handle occlusions and object interactions using learned cues.

Main Methods:

• Utilizing online target-specific metric learning for appearance cues.
• Employing coherent dynamics estimation for motion cues.
• Applying network flow optimization for tracklet association.
• Learning cue weights to manage challenging tracking situations.

Main Results:

• The proposed framework effectively associates track fragments.
• It demonstrates robustness against occlusions and object interactions.
• The method outperforms several state-of-the-art tracking approaches on public datasets.

Conclusions:

• The novel method enhances long-term multi-person tracking performance.
• Online learning of appearance and motion cues improves robustness.
• Network flow optimization provides an effective solution for tracklet association.