Disparity map algorithm for stereo matching process using local based method
The aim of Stereo Vision Disparity Map (SVDM) algorithm is to obtain the disparity map from two images. These input images have different viewpoints that corresponds with each other, forming a two-dimensional mapping of matching pixels and is known as disparity map. The SVDM algorithm can be categor...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English |
| Published: |
2022
|
| Subjects: | |
| Online Access: | http://eprints.utem.edu.my/id/eprint/26881/1/Disparity%20map%20algorithm%20for%20stereo%20matching%20process%20using%20local%20based%20method.pdf http://eprints.utem.edu.my/id/eprint/26881/2/Disparity%20map%20algorithm%20for%20stereo%20matching%20process%20using%20local%20based%20method.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The aim of Stereo Vision Disparity Map (SVDM) algorithm is to obtain the disparity map from two images. These input images have different viewpoints that corresponds with each other, forming a two-dimensional mapping of matching pixels and is known as disparity map. The SVDM algorithm can be categorized into local, semi global and global methods. Global method performs a matching process using global energy or a probability function over the whole image. This method involves high computational complexity and slow implementation. Therefore, it is not suitable for real-time applications. However, local method solves the matching problem via a local analysis and aggregating matching costs over a support region at each pixel in the images. The local method delivers fast execution and low computational requirement. Semi global method is the combination of both methods which make this method is more complex. However, the computation for the development of SVDM algorithm is more challenging especially for the images with complex scenes. There are several factors such as low texture region, repetitive patterns, illumination different, discontinuity, and occlusion. Hence, this thesis proposes a local-based SVDM algorithm that increases the accuracy on the complex scenes. The proposed SVDM algorithm involves four stages which starts from matching cost computation. At this stage, the proposed work uses the combination of Absolute Difference (AD) and Gradient Matching (GM) to produce the initial disparity map. Second stage, the Minimum Spanning Tree (MST) is utilised to remove noise from the initial disparity map. Then, the optimization stage uses a Winner-Take-All (WTA) strategy. The WTA strategy absorbs the minimal aggregated corresponding value for each valid pixel in disparity map. At the final stage, Bilateral Filter (BF) with Histogram Equalization and Weighted Median (WM) filter are proposed. These filters are capable to increase the accuracy and preserve the object edges. In this research, two standard online benchmarking database systems are used to measure the accuracy of the proposed algorithm. These systems are from the Middlebury Stereo for quantitative measurement and Karlsruhe Institute of Technology and Toyota Technological Institute (KITTI) for qualitative measurement. Once satisfactory results are obtained, there is demonstration on 3D surface reconstruction using disparity maps produced by the proposed SVDM algorithm. In conclusion, the proposed SVDM algorithm produces accurate results from the validation process. The final results are 7.55% for avg nonocc error and 10.6% for avg all error which are competitive with other established methods when compared in the standard benchmarking evaluation system from the Middlebury Stereo. The disparity map results on the complex scenes are also improved and fine quality of 3D surface reconstruction have been produced for both of the Middlebury and KITTI images. |
|---|