Improving accuracy, robustness and computational efficiency in 3D computer vision

This paper analyses the strengths and weaknesses of some of the most popular traditional and contemporary 3D vision techniques for accuracy, robustness and computational efficiency. A novel technique is proposed that extends traditional stereo vision algorithms using some of the previously identified techniques resulting in improved robustness, accuracy and computational efficiency. The new multi-scale temporally constrained stereo vision technique is then applied to a conventional stereo vision algorithm and the performance improvements demonstrated.