A computer vision algorithm for dynamic stereoscopic vision simultaneously addresses two conceptually distinct operations: tracking a moving object through successive frames and locating the object in space and time by measuring parallax displacement and other geometric properties of binocular vision. Based on a general smoothness assumption of minimal movement between successive frames, the algorithm demonstrates that the tracking and locating operations are treated in a more efficient manner simultaneously, rather than separately. A significant improvement over previous approaches is the algorithm’s ability to deal with moving objects that become fully or partly hidden from one or both eyes. The authors demonstrate the efficacy of their algorithm with a simulated walking stick-person and a digitized stereoscopic film of a toy boat. Their clearly described procedures are a useful contribution to the theory underlying developments in robot vision.