Question regarding the period argument in Tracker update method about norfair

[DiegoFernandezC]

Hi @romaintha

The approach for the period parameter to skip frames works a little differently. You can find a demo that uses period here.

The period argument is useful to skip frames and speed up the tracking by running the detector over a few frames.

The short answer is that, is not necessary to call multiple times the tracker_step() because you need to call the tracker.update() method on each frame and the tracker.update() uses the tracker_step() behind the scenes.

On each tracker.update() invocation Norfair runs the tracker_step() method, on this operation Norfair updates some attributes, for example, the hit_counter, and makes a prediction using the Kalman filter for the next position of each object.
As you noticed, this prediction should be executed as many times as the period value to avoid a big error when the detections arrive.
For this reason, the tracker.update() method allows you to not pass detections. This means that you have to run the tracker.update() on every frame. Then, when you get the detections run the tracker.update(detections=detections, period=period) passing the detections to the tracker. This allows Norfair to run the predictions despite not having the detections.

Norfair uses the period argument to compensate the hit_counter to the following tracker.update() invocations without detections.

The code should be looks like this:

for frame in video:
    detections = []
    if frame % period == 0:
        detections = model(frame)
    tracked_objects = tracker.update(detections=detections, period=period)

[romaintha]

I see. Thank you for your answer @DiegoFernandezC