Error Sources and Error Reduction in Gradient-Based Method with Local Optimization
暂无分享,去创建一个
The purpose of this study is to establish the technique for estimating optical flow with high accuracy and robustness using gradient-based method with local optimization. To obtain high accuracy, we should understand error sources and how to reduce the errors. We proposed error reduction techniques for gradient measurement error which are a spatiotemporal median filter to reduce sensor noise and a spatio-temporal derivative filter to estimate gradients of image function. The result shows that the spatio-temporal median filter can reduce the sensor noises very well, both of white noise and thermal noise of CCD camera. Furthermore, the best performance is achieved by the successive filtering of the Gussian filter and the spatio-temporal median filter. We also confirmed that estimation of partial derivatives of image function using the spatiotemporal derivative filter improved the accuracy of optical flow. The proposed methods are hopeful for the detection of optical flow with high accuracy and good robustness from image sequence.
[1] Tomoyuki Kudou,et al. Recovering 3D-Shape from Motion Stereo under Non-Uniform Illumination - A Vision System for Mobile Robot Carring an Illumination Source - , 1996, MVA.
[2] Hidetoshi Miike,et al. Evaluation of Body Motion by Optical Flow Analysis , 1997 .
[3] Joseph K. Kearney,et al. Optical Flow Estimation: An Error Analysis of Gradient-Based Methods with Local Optimization , 1987, IEEE Transactions on Pattern Analysis and Machine Intelligence.