Simple online and realtime tracking

The paper explores a pragmatic approach to multiple object tracking whose main focus is associating objects efficiently for online and real-time applications. Rather than introducing elaborate machinery, it uses a rudimentary combination of familiar components—the Kalman filter for motion prediction and the Hungarian algorithm for data association. A central observation is that detection quality strongly influences tracking performance: simply changing the object detector can improve tracking accuracy by up to 18.9%.

Despite its simplicity, the tracker achieves accuracy comparable to state-of-the-art online trackers, while running dramatically faster—updating at 260 Hz, over 20 times the speed of other state-of-the-art methods. This combination of competitive accuracy and very high speed makes it well suited as a practical baseline for real-time multi-object tracking.