Chang-Wu Fu, Shang-Hong Lai and Shyang Chang
Department of Electrical Engineering
National Tsing Hua University
Hsinchu, Taiwan, Republic of China
A method based on the spatial-temporal Hadamard transform is proposed to estimate 3-D translational motion parameters. In addition to the advantages of fast computation and easy implementation, this method has no restriction on the magnitude of displacement. For an image of size 2n¡Ñ2n, only 2n+1 coefficients need to be computed for translational parameters. Moreover, this method can be extended to the time-varying space-invariant (TVSI) illumination case. As a byproduct, the para- meters of rotation about the optical axis can also be estimated by the modified algorithm.
Furthermore, a modified version of Pentland's depth recovery algorithm can be incorporated into previous analysis to obtain the real-world translational motion under the constraint that only one camera is used. Finally, experimental results of this method and its comparison with the Fourier method are also illustrated.
Keywords: motion parameter estimation, hadamard transform, depth estimation, measurement from defocus, computer vision, dynamic scenes
Received August 9, 1991; revised May 5, 1992.
Communicated by Wen-Hsiang Tsai.