Development Of Stereo-Matching Algorithm Based On Adaptive Weighted Prediction
In computer vision field, one of most active topics in research is estimation of depth map which is Stereo Matching (SM) process and it’s also known as Stereo Vision Disparity Map (SVDM). The real challenge in SM is to get high accuracy of disparity map. Matching cost computation produces high noise...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English English |
| Published: |
2019
|
| Subjects: | |
| Online Access: | http://eprints.utem.edu.my/id/eprint/24618/1/Development%20Of%20Stereo-Matching%20Algorithm%20Based%20On%20Adaptive%20Weighted%20Prediction.pdf http://eprints.utem.edu.my/id/eprint/24618/2/Development%20Of%20Stereo-Matching%20Algorithm%20Based%20On%20Adaptive%20Weighted%20Prediction.pdf http://eprints.utem.edu.my/id/eprint/24618/ https://plh.utem.edu.my/cgi-bin/koha/opac-detail.pl?biblionumber=117088 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In computer vision field, one of most active topics in research is estimation of depth map which is Stereo Matching (SM) process and it’s also known as Stereo Vision Disparity Map (SVDM). The real challenge in SM is to get high accuracy of disparity map. Matching cost computation produces high noise of disparity map and input images contains low texture area and repetitive pattern also leads to high error on development of disparity map. Other than that, wrong information of each pixel of disparity map which can affect the accuracy of SVDM. Therefore, to overcome the causes of effected accuracy, new Stereo Matching Algorithm (SMA) based on Adaptive Weighted Bilateral Filter (AWBF) was introduced together with characterize the SMA based on quantitative and qualitative measurements and produced SMA performance were evaluate using standard taxonomy of SM. This thesis proposes an algorithm to handle the limitations. Firstly, pre-processing used Sobel Filter was added at initial step to compensate photometric distortion of input images. Then for matching cost, the proposed SMA combine one of matching cost method with some threshold adjustment to reduces the radiometric distortions. The Sum of Absolute Different (SAD) is the matching cost’s method and some threshold adjustment were used in this thesis where the SAD’s window size of 11 and threshold value of 0.8 was selected based on the experimental results. Secondly, to overcome low texture area and repetitive pattern limitation, this thesis present AWBF at cost aggregation stage where AWBF’s radius or window size of 19, spatial adjustment of 17 and disparity similarity value of 0.3 was selected. This process is introduced to preserve and improve the object boundaries. Finally, this thesis present Outlier Detection, window size of 5 for Median Filter and fill-in invalid disparity to utilize the last stage of SMA to handle the accuracy of disparity map in regions of occluded, repetitive and low texture. The experimental result on the proposed algorithm is able to reduce 17.4% of weighted average error for all and 9.62% of weighted average error for nonocc (nonoccluded) compared to others Stereo Matching Algorithm without the proposed framework. This framework experimental result was also compared with other methods which located in standard benchmarking dataset from the Middlebury. New SMA was developed in this thesis based on AWBF. The weighted average error of disparity map in both all and nonocc attribute are reduced based on the quantitative and qualitative measurements. Comparison of this framework continue in some of the state-of-the-arts algorithm in the literature and the results is outperformed based on the proposed algorithm experimental result. |
|---|