next up previous contents
Next: Depth and uncertainty Up: Correspondence Linking Algorithm Previous: Occlusions and visibility   Contents

Depth estimation and outlier detection

Care must be taken to exclude invalid disparity values or outliers from the chain. If an invalid disparity value is encountered, the chain is terminated immediately. Outliers are detected by controlling the statistics of the depth estimate computed from the correspondences. Inliers will update the depth estimate using a 1-D Kalman filter.


Marc Pollefeys 2002-11-22