Combined 3D object motion estimation in medical sequences

A method for the combined estimation of object´s motion and boundaries in 3D image sequences is proposed. Based on the object concept of MPEG-4 standard and on a preliminary segmentation, it partitions the images in 3 types of 3D blocks. Motion is first accurately determined on boundary blocks and initialize the estimation in the objects´ interior. This initialization is refined with a BBGDS algorithm to detect some disparities in the motion field. Synthetic experiments show that the accuracy is equivalent to classical block matching algorithms with exhaustive search but in less than half of the time. Moreover, the description of objects in terms of motion and shape is readily available.