Abstract:
Stereo Vision has conventionally been, and remains to be, one of the deeply examined
areas in computer vision. Stereo matching is used in numerous modern applications
including Robot navigation, augmented reality and automotive applications. In the absenceof radiometric variations stereo matching performs better than in the presence ofradiometric variations. In this thesis different stereo matching techniques are discussedand a new stereo matching technique, which compensates for radiometric variations, isproposed.
This work focuses on the improvement of the disparity map. It consists of three
main phases including pre-processing, stereo matching and post-processing. The input
images are firstly pre-processed with NMHE to compensate for radiometric variations.
Then different features are extracted from each view, which are insensitive to radiometricvariations, are employed in the state of the art local stereo matching algorithm.After this, the disparity map is post-processed to remove small artifacts from it.Experimental results shows that our proposed changes have improved the performanceof the state of the art local stereo matching algorithms and the robustness againstradiometric changes are improved
