A New Pre-Processing Technique For Computational Of Stereo Matching Algorithm

This paper presents a new composition of stereo vision algorithm for disparity map measurement from matching process. Stereo colour images obtained are consists with noises due to undesirable weather and illumination conditions due to it taken under inadequate or non-uniform light. The algorithm be...

Full description

Saved in:
Bibliographic Details
Main Authors: Kadmin, Ahmad Fauzan, Hamzah, Rostam Affendi, Abd Manap, Nurulfajar, Hamid, Mohd Saad
Format: Article
Language:English
Published: Research Publication 2021
Online Access:http://eprints.utem.edu.my/id/eprint/25540/2/AMSJ-2021-N02-06%281%29.PDF
http://eprints.utem.edu.my/id/eprint/25540/
https://research-publication.com/wp-content/uploads/2021/vol-10-n02/AMSJ-2021-N02-06.pdf
https://doi.org/10.37418/amsj.10.2.6
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Universiti Teknikal Malaysia Melaka
Language: English
Description
Summary:This paper presents a new composition of stereo vision algorithm for disparity map measurement from matching process. Stereo colour images obtained are consists with noises due to undesirable weather and illumination conditions due to it taken under inadequate or non-uniform light. The algorithm begins with pre-processing stage to enhance the colour image quality using combination of CLAHE, AGCWD and guided filter. Then, the matching cost computation are done using the Census Transform that has a strong advantage in radial distortion and brightness changes. The third stage will produce the aggregated cost from matching process utilizing fixed-window and guided filter technique. At the fourth stage; disparity optimization stage, the disparity map is optimized with a common local technique, Winner-Take-All (WTA). Then, for final stage, the process continues with post processing that is Left Right (LR) consistency checking. Weighted Median (WM) filter is that applied to secure the final disparity map for noise reduction and smoothening to the disparity map. Based on Middlebury Standard Benchmarking Dataset, the proposed algorithm has 23.35% accuracy for nonocc error and 31.65% accuracy for all error, which yields a better accuracy compared to some works in the evaluation dataset.