A performance analysis of stereo matching algorithms for stereo vision applications in smart environments

Abstract

Stereo vision is a subfield of computer vision that tends to an essential research issue of reproducing the three-dimensional directions and focuses for depth estimation. This paper gives a relative investigation of stereo vision and matching techniques, utilised to resolve the correspondence problem. The investigation of matching algorithms is done by the use of extensive experiments on the Middlebury benchmark dataset. The tests concentrated on an examination of three stereovision techniques namely mean shift algorithm (MSA), seed growing algorithm (SGA) and multi-curve fitting (MCF) algorithm. With a specific end goal to evaluate the execution, some statistics related insights were computed. The experimental results demonstrated that best outcome is attained by the MCF algorithm in terms of depth estimation, disparity estimation and CT. The presented MCF algorithm attains a minimum computation time (CT) of 2 s whereas the other MSA and SGA require a maximum CT of 8.9 s and 7 s, respectively. The simulation results verified that the MCF algorithm reduces the processing time in a significant way than the compared methods.

Plugging System of Rf Connector Based on the Fusion of Monocular and Binocular Vision

“…Because the existing relevant data set is small, it is not appropriate to use the deep learning method to train the neural network model. At the same time, considering the requirements of real-time detection, this paper selects the local matching algorithm as the detection algorithm [12]. The light intensity of the left and right images of the target area may be different, and the Census transform [13] can effectively eliminate the effect of light intensity differences, so the Census transform is used as the basis to obtain better results.…”