碩士 / 國立臺北科技大學 / 電腦通訊與控制研究所 / 89 / In this thesis, we have developed a binocular stereovision system to navigate an autonomous land vehicle. The main objective of the vision system is utilizing the similarity of the image features captured from left and right cameras to estimate world coordinate system. Hence, the system needs four steps:
1. Camera calibration
We employ eight corner points of the corridor to do the duty as calibration object and derive these intrinsic and extrinsic parameters by linear least-square method and orthonormal property.
2. Stereo corresponding
Due to the ALV following the wall in indoors with prior learning, we suggest a histogram-based approach to obtain both walls successfully. Moreover, we also suggest an edge linking approach based on grey prediction.
We suggest an improved approach to solve the stereo corresponding program. The algorithm combines the advantages of feature-based matching, region-based matching, and dynamical programming and cooperates grey relation analysis to solve the optimal correspondence point. Besides, if the image is very similar then it will obtain a wrong matching. Especially, the feature for stereo corresponding is intensity or region. Hence, we employ row vector to avoid this problem and improve the accuracy. Furthermore, we employ HSI color model, gradient magnitude and gradient direction as features to raise accuracy.
3. 3D Reconstruction
After the parameters and correspondence pairs are obtained by step 1 and 2, the 3D information of the point in world coordinate system can be solved by linear least-square method.
4. Indoor navigation
Generally, when human is walking, the people do not need to transform the local coordinate to global coordinate. Form the thought; we develop a navigation system of the ALV based on the trend like human and several experimental results have been conducted to verify the feasible about proposed approach for stereo corresponding.
| Identifer | oai:union.ndltd.org:TW/089TIT00652021 |
| Date | January 2001 |
| Creators | Wei-Yuan Cheng, 鄭惟元 |
| Contributors | Rong-Chin Lo, 駱榮欽 |
| Source Sets | National Digital Library of Theses and Dissertations in Taiwan |
| Language | en_US |
| Detected Language | English |
| Type | 學位論文 ; thesis |
| Format | 80 |
Page generated in 0.0152 seconds