Abstrakt

Advancement in Depth Estimation for Stereo Image Pair

Shabnam Hussain, Ravindra Modi

Many researchers invented ideas to compute depth estimation. In Hybrid Technique for Stereoscopic Depth Estimation there were some drawbacks associated with the optical flow estimation when it was done with novel disparity compensation step that the edges between the objects in disparity maps were not preserved very well which was caused in filtering step of optical flow analysis, related with desired smoothness. This research work presents an advance technique for the disparity map estimation from a stereo pair of images. The original concepts of proposal are: use of disparity estimation, a new hierarchical shape-adaptive block matching, optical flow analysis, novel occlusion detection and disparity extrapolation schemes. It is fairly insensitive to parameter variations, and it indicates its excellent robustness under noise. This method gives significantly smaller angular errors than previous techniques for optical flow estimation. The main advantages of this proposal are: low computational complexity, subpixel accuracy of depth estimation, operation across flat or untextured regions and good detection of occluded regions. The results of simulation are shown and compared different quality parameter of it’s by applying on various images.