Bipartite Matching

Hungarian algorithm for linking detections with predictions in tracking

• Links detections with predictions using distance metrics
• Uses IoU, Pixel, or 3D distances between boxes

Process

1. Calculate distances between boxes
2. Apply Hungarian algorithm for optimal matching
3. Set thresholds to handle missing/unsuitable matches

Edge Cases

• Missing prediction: Add dummy nodes
• No suitable match: Use cost threshold
• Unmatched boxes: Create new tracks

Output

• Matched pairs
• New tracks from unmatched detections
• Lost tracks from unmatched predictions

Hungarian Algorithm Demo