Efficient stereo image matching in man-made environments employing hough-based line descriptor

Abstract

Stereo image matching is one of the fundamental problems in the computer vision. Stereo image matching in the man-made structure is significant problem for the mobile device like robot, smart phone to 3D reconstruction of indoor or navigation. There exist several approaches to solve the stereo image matching like point, line segment features. The line feature has advantages in the man-made structure for its abundance. Efficiency in the stereo matching in man-made structure plays a importance role in mobile device due to limitation of processor. We propose the efficient stereo image matching in the manmade environments employing Hough-based line descriptor. First we xtract the line segments feature and extend it to prevent the fragmentation of line egments. Second, grouping of central line segments is determined. Degree number of losest line segments are included to the group of line segments. To reduce the effect from the noisy line segment, multi grouping scheme that the a several types of group is conducted. After that, line descriptor is calculated for the line segments in the group. The features in the Hough space is used for the descriptor. Finally each line segment is matched with the line segments between the stereo images. Experimental result shows that accuracy of correspondences of line segments is similar to the current research with huge amount of decreasing in computation time.