Tracking 2D Structures using Perceptual Organizational Principles

We investigate the role of perceptual organization in long image sequence analysis. Heretofore , the role of perceptual organization in computer vision has mainly been in static 2D image analysis. We exploit organizational principles to track extended 2D structures over time. We also impose organization on the tracks thus generated to extract vital motion information such as the focus of expansion and the center of rotation. We ooer an algorithm based on voting methods and graph theoretic operations for this purpose. The voting methods eeect a search among the image structures and the graph theoretic techniques are used to extract the feature tracks. The nature of the graph algorithm is such that the tracks are constructed by considering the global context of all the frames in the sequence and not by appending (local) two frame correspondences. We also present results on real images to demonstrate the viability of the algorithm. Abstract We investigate the role of perceptual organization in long image sequence analysis. Heretofore , the role of perceptual organization in computer vision has mainly been in static 2D image analysis. We exploit organizational principles to track extended 2D structures over time. We also impose organization on the tracks thus generated to extract vital motion information such as the focus of expansion and the center of rotation. We ooer an algorithm based on voting methods and graph theoretic operations for this purpose. The voting methods eeect a search among the image structures and the graph theoretic techniques are used to extract the feature tracks. The nature of the graph algorithm is such that the tracks are constructed by considering the global context of all the frames in the sequence and not by appending (local) two frame correspondences. We also present results on real images to demonstrate the viability of the algorithm. What is the original contribution of this work? In this paper we explore the role of perceptual organization in long image sequence analysis. Speciically, we exploit organizational principles to track 2D structures such as ellipses, circles, parallelograms, quadrilaterals, ribbons, and triangles, in time. The technique has the ability to handle the absence of features in intermediate frames due to occlusion or otherwise. We continue our exploration with the organization of the feature tracks themselves into meaningful patterns reeecting the underlying motion. Why should this contribution be considered important? Heretofore, the role of perceptual organization has mainly been in the …