Academic journal article International Journal of Electronic Commerce Studies

A Stereo Matching Algorithm Based on Adaptive Windows

Academic journal article International Journal of Electronic Commerce Studies

A Stereo Matching Algorithm Based on Adaptive Windows

Article excerpt

ABSTRACT

The aim of this paper is to develop a stereo matching algorithm based on adaptive windows for a stereo vision domain. This method retains the advantageous image processing speed of traditional methods, and proposes a means of decreasing the error rate, making it the best choice for application in real time systems. Depending on the characteristics of different regions, the proposed method provides a suitable window for stereo vision matching. The processing method is differentiated into disparity consistency, the disparity for a smooth region, the vote disparity between the 8-neighbors and the uniqueness of disparity. A different processing method is used in the lab with the sum of absolute difference (SAD), and the result is compared with a fixed window method; the result proves that this method improves on the SAD method, and yields more accurate depth information.

Keywords: Stereo Matching, Stereo Vision, Disparity, Adaptive Window, Sum of Absolute Difference

(ProQuest: ... denotes formulae omitted.)

1. INTRODUCTION

Currently, the field of robotics is receiving a great deal of attention; in this field machine vision is one of the most important sources of detection. If machine vision possesses high image identification ability, then robots can detect more diversified environmental changes and react suitably to those environmental changes. Machine vision has been developing for almost 40 years; many related studies and topics are being explored, with most of them roughly sorted and rearranged1, 2.

Stereo vision has become a very important topic of study, with special attention being paid to its application in machine vision. It is also a very important field in computer vision. There are several methods of conducting this type of research ranging for computer vision, using sensors like infrared rays, lasers, sonar, etc, to detect surrounding objects. Many of these sensors are able to acquire accurate three-dimensional coordinate information, but are very expensive. At present, camera and DV costs are decreasing, while resolution is continuously improving, and with the maturity of digital image processing and computer vision theory, stereo vision systems are being widely used for the measurement of distance in certain environments. A means of developing a relationship between three-dimensional coordinate systems and two-dimensional image coordinate systems for the measurement of three-dimensional stereo vision is an important study topic.

The purpose of camera calibration theory is to maintain horizontal direction consistency of overlapped images taken by several cameras. Stereo vision needs to be accurately calibrated before it can acquire more accurate three-dimensional information3; this process is a necessary pre-requisite to proficient stereo vision technique.

Stereo vision uses geometric relationships to determine the depth of a scene, on the basis of corresponding left and right images. Since the left and right images could possibly be affected by lighting to produce repetitive texture and occlusion2, if suitable and correct correspondence is not found, then errors will occur in the calculation of the scene depth. The correspondence of stereo images is therefore considered to be the most important problem in stereo vision.

Applying the matching relationship between left and right images can determine the reciprocal relationship between depth and disparity, and we can therefore acquire depth if we can determine the disparity value of each pixel. Disparity is mainly used to calculate the degree of difference of corresponding pixels in the left and right images; thereby, the relationship can be derived easily, with effective performance in real time systems. The means of determining correspondence can be differentiated into local 4and global5; the local method is widely applied to real time applications because it is less complicated than the global method. However, the local method cannot acquire high accuracy about occlusion, uniform texture and ambiguity portions. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.