Skip to content
BENIBIX Learn

Object Tracking

Also known as Multi-Object Tracking, Video Object Tracking

Following the same object through a video. Each one keeps its own ID.

Think of it like

It is like following one red balloon in a crowd. You keep your eye on it as it drifts.

Example

A traffic camera in Toronto tracks every car at a junction. Because each car keeps its ID, it counts 412 cars turning left in an hour.

Why it matters

Detection says what is in one frame. Tracking says where each thing went, which robots and cars need to predict moves.

Definition

Object tracking follows the same thing from one video frame to the next. First, a detector finds the objects in each frame. The tracker then links each box to the one it matches in the frame before. Each object keeps an ID number, so car 7 stays car 7 as it moves. This lets software count, measure speed or see where things go. The hard part is when objects cross, hide or look alike.
On the right, the car keeps ID 7 in every frame. That link is what tracking adds.

Common mistake

Thinking tracking is just detection on every frame. Without linking the boxes, you cannot tell one car from the next.

Key takeaways

  1. 1
    Link the frames. Match each box to the last one.
  2. 2
    Keep an ID. Car 7 stays car 7.
  3. 3
    Crossings are hard. Hidden or look-alike objects swap IDs.

Compare