Dense Depth Image Recovery Using Multi-Baseline Stereo Vision System

To eliminate fault correspondence, improve matching precision and recover dense depth image with high accuracy, a stereo matching method using multiple stereo pairs with various baselines was presented to obtain precise scene distance value without suffering from ambiguity. The constraints of epipolar geometry, uniqueness, smoothness and local normalizing process were imposed on the stereo images for uniquely finding corresponding points. The result shows that the errors of depth data are, on the average, less than 2 mm at distance between 1 5 m to 3 5 m away from the cameras.