| Summary: | 碩士 === 大同大學 === 電機工程學系(所) === 92 === The depth from stereo is one of the most active research topics in computer vision. In this work, we present a depth estimation algorithm suitable for implementation on a commodity PC. The proposed algorithm is based on the MML (Multiple MIP-map Levels) correlation-based algorithm presented by Yang and Pollefeys [1]. The weakness of the MML algorithm is that it fails to produce fair depth estimation around areas of low texture complexity and abrupt boundaries between distinct objects. By devising an effective and efficient method for background / foreground separation, we successfully tackle the MML problem and obtain fair good depth estimation over background objects as well as boundaries between background and foreground objects.
In our depth estimation system, firstly, we obtain a depth map by using the MML correlation-based method to perform the depth search with one half of the entire search range. The resulting depth map is sufficient for the estimation of background. After determining the background, we can exclude those background areas, and apply the rest of the depth search range to the MML correlation-based algorithm to update the depth for the remaining foreground areas. We successfully demonstrate the system by processing several stereo image pairs and obtain satisfactory experimental results.
|
|---|
