The idea is super easy to understand.

We need

  1. Object detector framework (YOLO, SSD, Faster RCNN etc.)

  2. Object motion model. The authors approximate the inter-frame displacements of each object with a linear constant velocity model. The state of each target is modeled as \[ \begin{equation} x_k = [u, v, s, r, \dot{u}, \dot{v}, \dot{s}] \end{equation} \] where $u$ and $v$ represent the horizontal and vertical pixel location of the centre of the target, while the scale $s$ and $r$ represent the scale (area) and the aspect ratio of the target’s bounding box respectively. Kalman filter is used to correct velocity components.

  3. The way to assign detections to existing targets The problem of assigning detected bounding boxes to existing objects is solved as linear assignment problem using the Hungarian algorithm.


  1. Simple online and realtime tracking