Depth Finding for Un-Calibrated Stereo Images by Area Based Matching Method
| Autor: | Ting-Cheng Feng, 馮鼎程 |
|---|---|
| Rok vydání: | 2007 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 95 Computer vision system has been widely used in our daily life. In general, two cameras are used to capture the stereo images of the scene around the computer vision system. These two images are 2-D only. It cannot directly provide the depth information for the objects located in the scene around the system. However, the relative position among the objects in each of the image is the same, due to they are captured for the same scene. In this thesis, an area based correlation method is proposed to find the depth information of the two stereo images. In strictly speaking, the propose method can be classified as a hybrid one. Firstly, one of the stereo images is segmented to get the rough area which covered the object. After the segmentation and find out the locations of the objects in one image, it is then overlapped onto the other unprocessed image to locate the starting searching region. Due to the two images are obtained from the same scene around the system and the two cameras are fixed with a relative position, the objects appeared on the two images must have a fixed relative position. So the matched area located on the unprocessed image corresponding to the segmented object in the processed image can easily be found by local area correlation matching method. By using this method, all of the corresponding objects between two images of a stereo vision can be matched, and this approach can effectively reduces the computational time and improving the matching quality. Accordingly, the depth of the stereo vision can be calculated. From the experimental results, it is shown that the error rate of the calculated depth information for the proposed approach is fewer than 10%. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|