Correlation-Based Visual Tracking Enhanced by Affine Motion Description

Since conventional correlation-based visual tracking algorithms are based on a single reference block, they often fail to track a target in the following situations: 1) a view of the target image changes, and 2) a part of the target image is occluded. This paper proposes a correlation-based visual tracking method which copes with these problems based on affine motion description. A reference block to be tracked is covered by multiple windows, which are partly overlapped each other. The motion of the reference block is approximated by an affine motion description by applying the least mean square method to motion vectors each of which is estimated from each tracking window. Based on the estimated parameters of affine transformation, the method detects the change in the view of the target image and updates the reference block. It also finds when and where the occlusion occurs, and keeps the reference block. Some results for real image sequences are given to show the validity of the meth...

• Type: Conference
• Year: 1996
• Where: MVA
• Authors: Yoshihisa Adachi, Minoru Asada, Takayuki Nakamura
• Keywords: Affine Motion Description, MVA 1996, MVA 2007, Reference Block, Target Image