I work on master's dissertation and I am looking for an method of calculating optical flow for pedestrian detection or pedestrian tracking based on optical flow.
So my approach is to detect pedestians at first. The detection algorithm must return coordinates of bounding box. Then, using bounding boxes I am going to track detected pedestrians.
I found out that OpenCV has calcOpticalFlowPyrLK method, but it uses individual pixels of frame for tracking, but not coordinates of bounding boxes.
So the questions are:
- Is there any method to resove pedestrian detection or pedestrian tracking tasks implemented?
- If yes, is there implementation of them on Python of C++?
- How effective are/is the methods/method?
P.S. If something isn't clear - just ask and I will make it clear. Thanks!
if you need something specific to pedestrians, you'll need to do literature research. there is a lot of research and implementations for the problem. getting that to run shouldn't be the scope of a thesis.
OpenCV has a whole module of trackers. they are not specific to pedestrians.
the dnn module also has some python samples for tracking (look for the siamese samples, "siam" in the file name) that likely are superior to those algorithms in the tracking module.
optical flow: there are sparse and dense methods. you decide what you need. pyrLK is a very old algorithm. if you want dense, use "DIS" optical flow. it's fast and good, also a more recent publication.