MULTI-SCALE MOTION ESTIMATION USING GRADIENT-BASED FILTERS
(0) Gradient-based algorithms play a vital role in motion estimation. A multi-scale motion estimation algorithm based on gradient filters was presented in this paper. In such an algorithm, the gradient filters were viewed as convolution of the discrete images with linear filters, and the motions between two images were estimated using data in many images. Thus, the accuracy of motion estimation is improved greatly. In combination with multi-scale pyramid approach, which the large pixels motions are converted to small motions, the estimator accuracy of such an algorithm is further improved. Experimental simulations show that the estimator performance performs very well for motion amplitudes larger than 1 pixel using this new algorithm, and estimator bias is less than 0.02 pixels for motions of noise images near 2 pixels. This represents a significant decrease in estimator bias compared to existing methods.
Motion Estimation Multi-scale Filter Gradient-Based
QINGHUA LU XIANMIN ZHANG
College of Mechanical Engineering, South China University of Technology, Guangzhou 510640, China
国际会议
开封
英文
372-380
2006-10-15（万方平台首次上网日期，不代表论文的发表时间）