Deep SORT

The improvements w.r.t. the SORT algorithm:

• Incorporates association metrics based on appearance features.

• Incorporates ID assignment to track individual objects across frames.

Steps

1. Detection and feature extraction

   • Using any detector

2. Apply Kalman filter

   • For the state prediction

   • Given the current position, velocity, and acceleration

   • Predicts the state of each object given the motion dynamics

3. Association:

   • Hungarian algorithm performs matching based on a cost matrix that considers:

     • Mahalanobis distance between bboxes (current and Kalman filters' prediction)

     • Cosine distance for appearance similarity

4. Track management (algorithm heuristics)

   • It confirms a track when is detected after several consecutive frames.

   • Age parameter to remove old tracks no longer existing.