During 1993-4, John spent a year in the Biorobotic Vision Group, part of the Centre for Visual Sciences at the Australian National University, working on motion-flow estimation and the segmentation of optical-flow maps derived from image-sequences. The objective was to distinguish objects in a scene, such as vehicles, pedestrians and background, using their observed relative movements. This involved segementing detected motions in a video-image sequence, and relating areas of similar motion in successive video frames. Optical flow was estimated using Srinivasan's generalised-gradient method.
John also investigated maximum a posteriori likelihood estimation as a possible means of improving the optical-flow data fed to the segementation algorithm.
This work was sponsored by Fujitsu, through Anutech.