Stereo and image matching on fixed size linear arrays

The authors present parallel techniques to implement two vision tasks; stereo matching and image matching using linear features as primitives. The implementations are performed on a fixed size linear array and achieve processor-time optimal performance. For stereo matching, they propose an O(Nn³/P) time algorithm on a P-processor linear array, where N is the number of line segments in one image, n is the number of line segments in a window determined by the object size, and P⩽n. The sequential algorithm takes O(Nn³) time. They also propose a partitioned parallel implementation of image matching with an O((nm/P+P)nm) time performance achieved on a P-processor linear array, where n is the number of line segments in the image, m is the number of line segments in the model and P⩽nm. This leads to a processor-time optimal solution for P⩽√nm. Previously known approaches to the image matching problem take O(n³m³) time