Announcement for Downloading full text filePlease respect the Copyright Act.
All digital full text dissertation and theses from this website are authorized the copyright owners. These copyrighted full-text dissertation and theses can be only used for academic, research and non-commercial purposes. Users of this website can search, read, and print for personal usage. In respect of the Copyright Act of the Republic of China, please do not reproduce, distribute, change, or edit the content of these dissertations and theses without any permission. Please do not create any work based upon a pre-existing work by reproduction, Adaptation, Distribution or other means.
URN etd-0911106-134538 Statistics This thesis had been viewed 2981 times. Download 1472 times. Author Chieh-Chih Chen Author's Email Address No Public. Department Communication Engineering Year 2005 Semester 2 Degree Master Type of Document Master's Thesis Language English Page Count 47 Title A STUDY OF DEPTH ESTIMATION ON A PC-BASED REAL-TIME STEREO VISION SYSTEM Keyword depth estimation stereo real-time real-time stereo depth estimation Abstract In this thesis, we construct a depth estimation algorithm suitable for real-time implementation on commodity PCs. The proposed method is based on the MML (Multiple Mip-map Levels) correlation-based algorithm. In the original MML method proposed by Yang and Pollefeys, the depth estimation does not fair well around areas of low texture complexity and abrupt boundaries between distinct object. Instead of constructing mip-map using box filter, we use Gaussian filter of different variance to construct the mip-map. This result in a more robust and accurate MML SSD (Sum of Square Difference).
We use Gaussian filters of smaller variances to reduce the degree of blurring in the higher mip-map level images whose image size is smaller. In the lower mip-map level images, we use Gaussian filters of larger variances to reduce the influence of noise. We generate the proposed MML SSD by combining the level 0 image of SD (Square Difference) image, the level 4 to level 6 mip-map images generated by the Gaussian filters of variance of 0.1, and the level 1 to level 6 images generated by the Gaussian filters of variance of 1.
We successfully demonstrate our system by processing several stereo image pairs and obtain satisfied experimental results.
Advisor Committee Jia-Ching Cheng - advisor
Shuenn-Shyang Wang - co-chair
Wei, Ching-Huang - co-chair
Files Date of Defense 2006-07-31 Date of Submission 2006-09-11