Employing Difference Image in Affine Tracking

G. Jing, D. Rajan, and C.E. Siong (Singapore)


Human tracking, affine tracking, difference image.


Human tracking is a well developed research area in computer vision and various algorithms have been brought up. Some researchers see tracking as a iteratively optimiza tion process in each frame to find the optimum transfor mation parameter that minimize the distance between the transformed region and the object model. The most of ten used transformation are translations and affine warps. In this paper, using affine tracking as example, we com bine difference image between consecutive image frames into the optimization process, i.e the difference image pro vides several starting points of gradient decent to search for the object, rather than only using one starting point to find the candidate region. In this way our tracker have more discriminating powers when occlusion happens without re sorting to sampling methods such as particle filter. Experi mental results demonstrate the effectiveness of our method.

Important Links:

Go Back