Multi-Object Tracking (MOT) is the task of estimating the trajectory of multiple unique objects in a video — assigning a unique ID to each detected object and maintaining that ID even as objects cross paths, are occluded, or move erratically.

What Is MOT?

• Paradigm: Detection-by-Tracking vs. Tracking-by-Detection.
• Standard Pipeline:

1. Detect objects in current frame (YOLO). 2. Extract features (Re-ID embedding + Motion/Kalman Filter). 3. Associate with existing tracks (Hungarian Algorithm).

• Metric: MOTA (Multiple Object Tracking Accuracy), IDF1.

Why It Matters

• Traffic Monitoring: Counting distinct cars, not just detections per frame.
• Crowd Analysis: Tracking flow of people in public spaces.
• Retail: Tracking customer paths through a store ("Customer Flow").

Key Failure Mode: ID Switch. When two people cross paths and the tracker swaps their IDs.

Multi-Object Tracking is converting perception into identity — turning raw detections into persistent, trackable entities.