Collaboration on Waltz Labels can Achieve Qualitative Stereo Vision

Abstract

Stereo vision requires calibration that can be hard to achieve or guarantee, so we propose a Qualitative Stereo Vision approach based on logical reasoning about edges detected from cameras/robots having different points of view. The proposed technique builds upon the qualitative reasoning of Waltz filtering on edges of a 2D image, which is extended to reasoning about edges from multiple images with different points of view. An assumption is that edges and vertexes that appear in different images can be identified based on their features. We find that a consistent spatial interpretation of a scene, classifying occluding, convex, and concave edges, can be obtained by extending an intersection of labels for common edges with a ``convex'' option in those cases when available ``occlusions'' semantics do not match for corresponding edge sides. The power of the proposal in generating qualitative stereo vision is illustrated with case studies.