Counting human movement of exercise using ROI or pre-set passing line with OpenCV

Question

I'm currently doing my final year Degree in computing studies.

The project I'm working on is a video of a webcam or computer webcam to have 2 pre-defined ROI (Region of Interest) on top and bottom of the video. It is to detect a person when he does a simple push up whether he reaches the area of both top and bottom ROI by detecting the person's head when he enters the ROI. I totally have very less experience in programming and really need guide and help.

During my research I've come across a journal/research at: Pattern Recognition Letters 33 (2012) 438–445. I've downloaded and gone thru it but can't seem to find its code or truly understand it. I got access with my University account at sciencedirect.com.

2nd idea I had was after viewing a video in Youtube: Traffic Counting based on OpenCV... I was very interested by that video. I was thinking if I could set 2 lines like he did and detect simple human motion up and down doing a simple accurate pushup exercise I would be so much happy but I can't seem to get contact or ask for any information.

I really need advice and help because I am really lost.

Answer 1

This might help partially -

You can, possibly track the person's head from his hair color(usually black), so can be distinguished from the background. This tracking just needs a simple RGB Filter like ( r<val2 && g<val4 && b<val3). Then, you can calibrate the program to respond when the centroid of that hair-region crosses your desired lines/boundaries.

Answer 2

To track a human you have many options. If you could place a Kinect camera for example and track the skeleton of the person, you could obtain much more statistics about the behavior. This way, you could also get rid of the burden of segmenting the person. OpenCV already has Kinect integration.