Body Tracking Algorithms without RGB?

asked 2013-12-26 14:49:47 -0600

I have an Xtion Pro, which provides a depth sensor but not an RGB sensor. I would like to perform body tracking using OpenCV, and I only want to detect only upper body (torso and up). I have one moving camera at a fixed height (chest-level to the person). I have researched several ways of doing so such as: HOG, LBP, HAAR, latent SVM, and Kalman tracking. I would like to find out however if these methods can be implemented in an algorithm without an RGB sensor. If so, which methods would you suggest and why? If not, are there other methods using only depth sensing?

Thank you for your time!

edit retag flag offensive close merge delete


interesting idea !

to my knowledge, HOG, LBP, HAAR, are computed on grayscale images, not rgb.

still, a depth sensor might have less than 255 intensity values.

berak gravatar imageberak ( 2013-12-26 15:12:44 -0600 )edit

When you have depth data there are much simpler methods you should try before using heavy classification techniques. Thresholding to max admissible depth (distance of user from sensor) and then try some blob analysis would be my first approach.

Pedro Batista gravatar imagePedro Batista ( 2013-12-27 04:44:18 -0600 )edit