Object Discovery through Motion, Appearance and Shape
Tristram Southey, James J. Little · 2006
In this thesis we examine the problem of Object Discovery, the autonomous acquisition of object models, using a combination of shape, appearance and motion. We propose a new technique for detecting rigidly moving objects and constructing models of their appearance and shape called the ODMAS (Object Discovery through Motion, Appearance and Shape) system. Our technique is a multi-stage approach. First, a stereo camera is used to find a sequence of images and shape maps of a given scene. Then the scene is oversegmented using normalized cuts based on a combination of shape and appearance. SIFT image features are matched between sequential pairs of images to identify groups of moving features and the three dimensional location of these moving features in the scene is identified using the shape map from the stereo camera. The moving features are then further individuated into objects by identifying groups where the motion of the features is rigid. These rigid feature groups are used to determine which regions in the segmentation of the scene correspond to objects,