Video Segmentation (kinect + openni + opencv) 2

I added the skeleton information, which can help me to extract human body from the video.

In the old version of NITE, before the skeleton calibration, it should detect Psi pose first. But after the version 1.5, this step can be omitted. That makes the process more easy.

Firstly, I use the skeleton information to draw the skeleton, and set these area as foreground. Then, I set the average depth of the skeleton as depth threshold. So, I can distinguish foreground and background roughly.

Secondly, I use the color information to do border matting. The main idea is based on GraphCut algorithm.

 

Figure 1. binary image, depth image and skeleton image

When the difference of color between foreground and background is large, the edges will be good.

However, when the foreground's color is similar with background's, or the unknown area of depth image is large, there will be some mistakes.

Figure 2. partial enlarged detail

 

转载于:https://www.cnblogs.com/ytabd6/archive/2013/04/22/3035016.html