Inference of Segmented , Volumetric Shape from Intensity Images

reproduce and distribute reprints for governmental purposes notwithstanding any copyright notation hereon. Abstract We present a method to infer segmented and volumetric descriptions of objects from intensity images. Our descriptions are in terms of Generalized Cylinders (GCs). We use three weakly calibrated images taken from slightly different viewpoints as input, where the object is only partially visible. There are parts of the object whose surfaces are facing away from the camera and parts are self occluded. Arriving at volumetric descriptions in these cases, with only partial data, requires the development of strong inference rules. These rules are based on the local properties of GCs.We first detect groups in each image based on proximity, parallelism and symmetry. The groups are matched and their contours are labelled as " true " and " limb " edges. We use the information about groups and the label associated with its contours to recover visible surfaces. We then use local properties of GCs to obtain the position of the GC axis and the cross sections to make a volumetric inference. The final descriptions are volumetric and in terms of parts. We demonstrate results on real images of moderately complex objects with texture and shadows.