hit_inertia_min and initialization_delay meaninng and behaviour about norfair

[aguscas]

When the hit_inertia value of a TrackedObject instance reaches hit_inertia_min, you will find that object in the tracked_objects list of the tracker, but will not be returned when you update your tracker with your current detections until the object finish initializing. The object is being initialized until the hit_inertia value exceeds hit_inertia_min by initialization_delay, after that, the objects has been initialized and will be assigned an id value, and you will see it in the list that is returned when you update the tracker with your detections.

I think the output you showed here corresponds with the tracked_objects list of the tracker, instead of the list returned by the tracker when you update.

[dinarkino]

@joaqo thank you! I see, if the counter starts with hit_inertia_min + 1, then everything fall into place. Yeah, an update of docs will be much appreciated.

[joaqo]

Great! Yeah, the docs are in need of a lot of work, we'll fix them soon 😀

[joaqo]

Closing due to inactivity.

[dinarkino]

When the hit_inertia value of a TrackedObject instance reaches hit_inertia_min, you will find that object in the tracked_objects list of the tracker, but will not be returned when you update your tracker with your current detections until the object finish initializing. The object is being initialized until the hit_inertia value exceeds hit_inertia_min by initialization_delay, after that, the objects have been initialized and will be assigned an id value, and you will see it in the list that is returned when you update the tracker with your detections.

I think the output you showed here corresponds with the tracked_objects list of the tracker, instead of the list returned by the tracker when you update.

Thank you for the replay and sorry for the late answer. I see. You say that there is a difference between tracked_objects inside the tracker and the output of the tracker.update. But still, the behaviour of counters is not clear to me. First of all, here I printed the values of the output of tracker.update. My code:

hit_inertia_min = 5
hit_inertia_max = 10
initialization_delay = 3
visualize = True
    
tracker = Tracker(
        distance_function=euclidean_distance,
        distance_threshold=max_distance_between_points,
        hit_inertia_min=hit_inertia_min,
        hit_inertia_max=hit_inertia_max,
        initialization_delay=initialization_delay,      
    )

for i in range(20):      
    # here I use a message from ROS but never mind it is just a fixed bbox 
    # which I then convert to Norfair format
    test_box = BoundingBox()
    test_box.xmin = 514
    test_box.ymin = 317
    test_box.xmax = 560
    test_box.ymax = 355

    msg = BoundingBoxes()
    msg.bounding_boxes = [test_box]

    detections = msg.bounding_boxes
    detections = ros_detections_to_norfair_detections(detections)
    
    # no boxes from some point
    if i > 10:
        msg = BoundingBoxes()
        detections = msg.bounding_boxes

    tracked_objects = tracker.update(detections=detections)  
    
    print(f'Update number: {i}')
    print(f'tracker.tracked_objects: {tracker.tracked_objects}')
    print(f'tracker.update(detections=detections): {tracked_objects}')
    print('____________')

I don't know how to access these two different entities but for tracker.tracked_objects and tracker.update(detections=detections) output is the following:

Update number: 0
tracker.tracked_objects: [Object_None(age: 0, hit_counter: 6, last_distance: None, init_id: 15)]
tracker.update(detections=detections): []
____________
Update number: 1
tracker.tracked_objects: [Object_None(age: 1, hit_counter: 7, last_distance: 0.00, init_id: 15)]
tracker.update(detections=detections): []
____________
Update number: 2
tracker.tracked_objects: [Object_None(age: 2, hit_counter: 8, last_distance: 0.00, init_id: 15)]
tracker.update(detections=detections): []
____________
Update number: 3
tracker.tracked_objects: [Object_1(age: 3, hit_counter: 9, last_distance: 0.00, init_id: 15)]
tracker.update(detections=detections): [Object_1(age: 3, hit_counter: 9, last_distance: 0.00, init_id: 15)]
____________
Update number: 4
tracker.tracked_objects: [Object_1(age: 4, hit_counter: 10, last_distance: 0.00, init_id: 15)]
tracker.update(detections=detections): [Object_1(age: 4, hit_counter: 10, last_distance: 0.00, init_id: 15)]
____________
Update number: 5
tracker.tracked_objects: [Object_1(age: 5, hit_counter: 11, last_distance: 0.00, init_id: 15)]
tracker.update(detections=detections): [Object_1(age: 5, hit_counter: 11, last_distance: 0.00, init_id: 15)]
____________
Update number: 6
tracker.tracked_objects: [Object_1(age: 6, hit_counter: 10, last_distance: 0.00, init_id: 15)]
tracker.update(detections=detections): [Object_1(age: 6, hit_counter: 10, last_distance: 0.00, init_id: 15)]
____________
Update number: 7
tracker.tracked_objects: [Object_1(age: 7, hit_counter: 11, last_distance: 0.00, init_id: 15)]
tracker.update(detections=detections): [Object_1(age: 7, hit_counter: 11, last_distance: 0.00, init_id: 15)]
____________
Update number: 8
tracker.tracked_objects: [Object_1(age: 8, hit_counter: 10, last_distance: 0.00, init_id: 15)]
tracker.update(detections=detections): [Object_1(age: 8, hit_counter: 10, last_distance: 0.00, init_id: 15)]
____________
Update number: 9
tracker.tracked_objects: [Object_1(age: 9, hit_counter: 11, last_distance: 0.00, init_id: 15)]
tracker.update(detections=detections): [Object_1(age: 9, hit_counter: 11, last_distance: 0.00, init_id: 15)]
____________
Update number: 10
tracker.tracked_objects: [Object_1(age: 10, hit_counter: 10, last_distance: 0.00, init_id: 15)]
tracker.update(detections=detections): [Object_1(age: 10, hit_counter: 10, last_distance: 0.00, init_id: 15)]
____________
Update number: 11
tracker.tracked_objects: [Object_1(age: 11, hit_counter: 9, last_distance: 0.00, init_id: 15)]
tracker.update(detections=detections): [Object_1(age: 11, hit_counter: 9, last_distance: 0.00, init_id: 15)]
____________
Update number: 12
tracker.tracked_objects: [Object_1(age: 12, hit_counter: 8, last_distance: 0.00, init_id: 15)]
tracker.update(detections=detections): [Object_1(age: 12, hit_counter: 8, last_distance: 0.00, init_id: 15)]
____________
Update number: 13
tracker.tracked_objects: [Object_1(age: 13, hit_counter: 7, last_distance: 0.00, init_id: 15)]
tracker.update(detections=detections): [Object_1(age: 13, hit_counter: 7, last_distance: 0.00, init_id: 15)]
____________
Update number: 14
tracker.tracked_objects: [Object_1(age: 14, hit_counter: 6, last_distance: 0.00, init_id: 15)]
tracker.update(detections=detections): [Object_1(age: 14, hit_counter: 6, last_distance: 0.00, init_id: 15)]
____________
Update number: 15
tracker.tracked_objects: [Object_1(age: 15, hit_counter: 5, last_distance: 0.00, init_id: 15)]
tracker.update(detections=detections): [Object_1(age: 15, hit_counter: 5, last_distance: 0.00, init_id: 15)]
____________
Update number: 16
tracker.tracked_objects: [Object_1(age: 16, hit_counter: 4, last_distance: 0.00, init_id: 15)]
tracker.update(detections=detections): [Object_1(age: 16, hit_counter: 4, last_distance: 0.00, init_id: 15)]
____________
Update number: 17
tracker.tracked_objects: []
tracker.update(detections=detections): []
____________
Update number: 18
tracker.tracked_objects: []
tracker.update(detections=detections): []
____________
Update number: 19
tracker.tracked_objects: []
tracker.update(detections=detections): []

We get tracked_objects not after hit_inertia_min, but at the first detection appearance. Then we get objects in the output of tracker.update at third detection appearance, not at hit_inertia_min + initialization_delay. And thirdly, why do we have hit_counter = 6 right at the beginning. Maybe I'm doing something wrong, I will be glad for the help.

[joaqo]

Hey @dinarkino I think the confusion is because each tracked object starts with hit_inertia_min + 1 as its hit_counter. And then if it gets over initialization_delay + 1 (if you set initialization delay) it starts counting as a real object. We know that our docs are not very good with regards to this, and we are going to improve them, and make the api much simpler. Sorry for the inconvenience.