Abstract
Stereo matching is a method to obtain the depth information from images and is one of the most important issues in the field of machine vision. Contrary to the conventional stereo matching algorithms that involve two or more wellcalibrated cameras, the depth extraction scheme using a lens array is compact and requires no calibration since only one camera is involved in its setup. In this paper, we present a novel depth extraction algorithm using a lens array. The proposed method rearranges horizontal positions of the pixels from the collection of the elemental images to form subimages horizontally leaving the vertical positions of the pixels unchanged. On this rearranged images, we apply a correlation-based multiple-baseline stereo algorithm in properly modified form. The main feature of the proposed method is its ability of exact depth extraction from the extremely periodically patterned object scenes. Additionally, the proposed method enlarges the available depth range due to the reverse dependency of the disparity on the depth between the elemental image and the sub-image. We prove our idea by applying our method on the object scene generated by a computer. The simulation result shows the proposed method extracts precise depth information from the scene of the object with a periodic texture pattern.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call