Stereo Imaging for 3D Scene Reconstruction

Stereo Vision plays an important role in computer and robot vision. It is similar in concept to human binocular vision useful for three dimensional scene reconstruction. This research presents a novel stereo vision technique with a fast left-right searching for stereo matching. The system has been furnished with the disparity estimation of the image sequences using the sum of absolute differences (SAD) and sum of squared differences (SSD) methods and reconstruction of the 3D scene from the obtained disparity map. This research presents three major contributions: (i) the implementation of a fuzzy filtering to reduce additive noises from the stereo images, (ii) the development of a fast stereo correspondence estimation method called ¿left-right searching¿ that increases matching accuracy and reduces computation time substantially with no appreciable degradation of accuracy, and (iii) the development of a new stereo based model for 3D scene reconstruction. Experimental results indicate that the proposed stereo vision system is capable of matching correct pixels more than 70% and it reduces the window cost computation time of about 44% compared to other traditional techniques.