Motion Segmentation in Long Image Sequences

Abstract

Long image sequences provide a wealth of information, which means that a compact representation is needed to efficiently process them. In this paper a novel representation for motion segmentation in long image sequences is presented. This representation ‐ the feature interval graph ‐ measures the pairwise rigidity of features in the scene. The feature interval graph is recursively computed, making it a compact representation, and uses an interval model of uncertainty. The feature interval graph forms the basis for new algorithms for motion segmentation and occlusion analysis. Results of these algorithms are presented on synthetic and laboratory scenes. The analysis of long image sequences is a growing area of research in computer vision. Long sequence analysis is important for visual surveillance, mobile robotics, and other areas where a dynamic scene is observed over a long period of time. Long image sequences provide a wealth of information, but this raises the problem of efficient representation. It is not feasible to store an entire sequence, and so a compact representation is needed which can be efficiently computed, and does not grow with the length of the sequence. In this paper the particular problem of motion segmentation in long image sequences is addressed. In motion segmentation the goal is to cluster the scene, or features extracted from it, into regions having a common motion. The clusters correspond to independently moving objects in the scene and so are useful for tracking and navigation. Section 2 presents some previous approaches to the task of motion segmentation in long image sequences. These approaches make use of the Kalman filter [2], which provides a compact and robust representation of information gathered from the image sequence. The Kalman filter, however, relies on a Gaussian model of uncertainty, and assumes that the variances of these distributions are known. In Section 3 a new representation for motion segmentation is presented. This representation, the feature interval graph, measures pairwise rigidity information directly and shares many of the advantages of the Kalman filter, but requires less a priori knowledge. The feature interval graph is then used to develop new algorithms for motion segmentation and the analysis of partial occlusions in the scene in Sections 4 and 5. An analysis of these algorithms is given in Section 6, followed by some concluding remarks.