1) it is necessary to give the precision especially for the sequences where there is overlapping of the similar objects.
2) percentage of the pixels of the tracked object that are well calcified
3) the method must also be tested in the case of noisy sequences while giving the percentage of pixels well classified
...
Simple: Test the object tracking method and check how far you are along time...
Mostly for object tracking the measure accuracy is not used very common, another measure which is in consideration is the IoU, intersection over union as you are tracking it should be determined that how much region of your target is in intersection with the bounding box.
Which technique used for tracking? If it is for overlaped may be defined..
Window based method can be used, with all objects window different colored so easy to track overlapped object.
I am going to assume that you know the ground truth bounding boxes of the objects. In this case, you could use metrics such as mean average precision, intersection over union, or euclidean distance between the bounding box centers averaged over time.
I am not sure if you already looked at this github project, I found it to answer your question to some extent: https://github.com/rafaelpadilla/Object-Detection-Metrics
- Badges
- Science topic