| Author | Wu, Xiao-wen |
| Call Number | AIT Diss. no.CS-97-1 |
| Subject(s) | Image processing--Digital techniques
|
| Note | A dissertation submitted in partial fulfillment of the requirements for the degree of Doctoral Technical Science |
| Publisher | Asian Institute of Technology |
| Abstract | Stereo matching is a fiindamental process in photogrammetry and computer vision. In the last four decades many attempts have been made to automate it. Until now, a fully automated and reliable matching method seems not exist yet. The objective of this research is to develop a new matching method and analyze the performance of this algorithm. In this research, the linear array CCD cameras were chosen as the imaging device. Linear array CCD cameras have been used extensively in many applications such as, industrial measurements, quality inspection and air survey. They are especially suitable in the environment in which inherent movement between object and the cameras exists. Due to the complexity of the stereo matching problem1 it is well known that a single feature or single resolution approach cannot lead to a general solution. In recent years, using multi-resolution or multi—feature matching to solve the matching problem has become two promising research directions. In'this research, these two approaches were combined into a multi-resolution and multi-feature matching based on the wavelet transform decomposition. The original images were decomposed by the wavelet transform into several filtered images which contain information in difi‘erent orientation and fiequency ranges. A multi-feanire matching was carried out in each level and the results fi'om different levels were combined to form the final results. It was shown that better results for the three test cases have been achieved compared with using only single feature and single resolution matching. Another advantage of this approach is that the wavelet transform provides a unique solution for both feature extraction and feature matching. The feature extraction is accomplished automatically during the wavelet transform decomposition. To further improve the matching results, a three line scanner (TLS) image matching algorithm has been proposed. The redundancy in the triplet images can be used to improve the matching results. The disparity constraint of the TLS images has been investigated and exploited to refine the initial disparity estimation1 which is produced by a stereo image matching. A model based interpolation has been applied to generate the final results. The effectiveness of our method with TLS, was demonstrated by comparing the results with those of matching only two line scanner images. It was shown that better results were achieved with TLS in the three test eases. Finally, the disparity space image was used to integrate the two algorithms (the multi- feature and multi-resolution matching. and the multi-image matching). The matching results fer the three test cases showed the best as compared with other methods. In addition, a parallel implementation of the matching algorithm was tested with two to six workstations in order to speedup the computation. The results of the parallel matching showed satisfied speedup has been obtained. |
| Year | 1997 |
| Type | Dissertation |
| School | School of Engineering and Technology (SET) |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Computer Science (CS) |
| Chairperson(s) | Honda, K. |
| Examination Committee(s) | Murai, S.;Sadananda, R.;Yulu, Qi, |
| Degree | Thesis (Ph.D.) - Asian Institute of Technology, 1997 |