Mastering Computer Vision with TensorFlow 2.x

Centroid-based tracking

As the name suggests, centroid-based tracking involves tracking the centroid of the image cluster developed using thresholding. At initialization, IDs are assigned to the bounding box centroid. In the next frame, the IDs are assigned by looking at the relative distance between two frames. This method works when objects are far apart but does not work when objects are very close to each other.