Object Tracking Visualizer
Run the Object Tracking Visualizer Fullscreen
Edit the Object Tracking Visualizer with the p5.js editor
About This MicroSim
This visualization demonstrates multi-object tracking (MOT) - the problem of maintaining consistent identities for multiple moving objects across frames. This is essential for autonomous vehicles to track pedestrians, vehicles, and cyclists.
Embedding
You can embed this MicroSim in your website using:
1 2 | |
Features
- Multiple Objects: Colored bounding boxes with unique track IDs
- Predictions: Dashed boxes show where tracks are expected to appear
- Detections: Red boxes show noisy sensor measurements
- Associations: Yellow lines connect predictions to matched detections
- Track Trails: Historical positions showing object paths
- Track Management: New tracks created, lost tracks deleted
Key Concepts
The Tracking Pipeline
- Prediction: Use motion model to predict where each track will be
- Detection: Generate (noisy) measurements of objects
- Association: Match predictions to detections
- Update: Refine track states with matched detections
- Track Management: Create new tracks, delete lost tracks
Data Association
The key challenge is matching detections to tracks: - Gating: Only consider detections within a distance threshold - Cost Matrix: Compute distance between all prediction-detection pairs - Hungarian Algorithm: Find optimal one-to-one assignment
Association Metrics
| Metric | Description |
|---|---|
| IoU | Intersection over Union of bounding boxes |
| Mahalanobis | Distance accounting for uncertainty |
| Euclidean | Simple distance between centers |
Lesson Plan
Learning Objectives
- Understand the multi-object tracking pipeline
- Recognize the role of prediction in data association
- Apply tracking concepts to handle missed detections
Activities
- Step Through: Click "Step" to see each predict-associate-update cycle
- High Noise: Increase detection noise, observe association errors
- High Miss Rate: Increase miss probability, see tracks use predictions
- Track Creation/Deletion: Observe new tracks spawn, lost tracks disappear
Assessment Questions
- Why do we predict before associating detections?
- What happens when a detection is missed for several frames?
- How would you handle objects that cross paths?
References
- SORT: Simple Online Realtime Tracking
- DeepSORT
- Chapter 15: Autonomous Systems and Sensor Fusion