Though background subtraction is common computer vision task, in 2D image processing its difficult to correctly differentiate the foreground from background. By observing 2D image our brain visual system can sense and differentiate the foreground and background easily but if we want build the same system on computer, its really challenging. Our surrounding world is 3D, but due to limitations of technology in most of the computer vision application we use 2D only.
The major drawback of 2D image is its gives intensity and colour values only. Though 2D images gives depth cues, correctly differentiating background and foreground is not possible with 2D image processing. If we consider 3D image, one more dimension is added over 2D image. That third dimension is nothing but depth of each pixel. If we get depth for each pixel, we can easily segment the scene in foreground and background on the basis of depth.
To demonstrate foreground-background subtraction in 3D, I used Kinect sensor which gives me depth for each pixel. On the basis of depth value, I have segmented the scene in foreground and background. I capture the scene at a distance from 1.5 meter to 5 meter from sensor. For quick demo, go through video :
No comments:
Post a Comment