Abstract:

 As the wavelet transform has outstanding local characteristics both in temporal and spatial domains,
moving objects are extracted based on background subtraction using wavelet multi-scale transform. According to the
relative spatial position of pointing arm finger in a 2D image, a method for the extraction of pointing arm region is
developed based on fusion of threshold segmentation and color information. In processing the correspondence problem
of finger tip pointing, a stereo matching technique is used based on geometric constraints among the pointing arm
finger tip, central axis of the pointing arm, and the corresponding epipolar line. Reverse correspondence is adopted
to ensure validity of the processed correspondence. Experimental results show that the described algorithm is robust
and 3D information of the pointing arm can be determined efficiently.

Key words: movement localization, video tracking, object segmentation, stereo vision, correspondence