Applying Recursive EM to Scene Segmentation

In this paper a novel approach for the interdependent task of multiple object tracking and scene segmentation is presented. The method partitions a stereo image sequence of a dynamic 3-dimensional (3D) scene into its most prominent moving groups with similar 3D motion. The unknown set of motion parameters is recursively estimated using an iterated extended Kalman filter (IEKF) which will be derived from the expectation-maximization (EM) algorithm. The EM formulation is used to incorporate a probabilistic data association measure into the tracking process. In a subsequent segregation step, each image point is assigned to the object hypothesis with maximum a posteriori (MAP) probability. Within the association process, which is implemented as labeling problem, a Markov Random Field (MRF) is used to express our expectations on spatial continuity of objects.