OpenCV 4 with Python Blueprints - Second Edition

Tracking detected objects

Once we can successfully detect objects in each frame, we can track them by associating detections between frames. As mentioned previously, in this chapter, we are using the Sort algorithm for multiple-object tracking, which stands for Simple Online and Realtime Tracking.

Given sequences of multiple bounding boxes, this algorithm associates the boxes of sequence elements and fine-tunes the bounding box coordinates based on physical principles. One of the principles is that a physical object cannot rapidly change its speed or direction of movement. For example, under normal conditions, a moving car cannot reverse its movement direction between two consequent frames.

We suppose that the detector annotates the objects ...