Review of Stereo Vision Algorithms: From Software to Hardware

Abstract

Stereo vision, resulting in the knowledge of deep information in a scene, is of great importance in the field of machine vision, robotics and image analysis. In this article, an explicit analysis of the existing stereo matching methods, up to date, is presented. The presented algorithms are discussed in terms of speed, accuracy, coverage, time consumption, and disparity range. Towards the direction of real-time operation, the development of stereo matching algorithms, suitable for efficient hardware implementation is highly desirable. Implementations of stereo matching algorithms in hardware for real-time applications are also discussed in details.

Texture-Aware Dense Image Matching Using Ternary Census Transform

“…Taking traditional area-based matching as an example, the matching cost is the square difference of pixel intensities, matching cost aggregation is generally accomplished using the Normalized Correlation Coefficient (NCC) in a rectangular window, the disparity computation is carried out through a local winnertake-all (WTA) operation in a small search window and screening with a simple threshold, and the final step typically involves sub-pixel interpolation and other post processing. DIM methods are classified as either local or global, distinguished by the use of smoothness constraints and the inference of disparities (Scharstein and Szeliski, 2002;Lazaros et al, 2008). The former depends only on local intensity values, while global reasoning is generally used in the latter for disparity optimization.…”

“…Textureless areas will cause ambiguities in determining matching cost and therefore magnify errors from the random noise of pixel intensities. Because of the difficulties to obtain local optimum in matching of textureless areas, DIM methods will generally fail in this area without smoothness constraint (Lazaros et al, 2008). The simplest strategy to exploit smoothness is to aggregate matching cost in a larger window (Xin et al, 2012), as the aggregation window implicitly uses the assumption of the same disparity value (Yang, 2015).…”

An Improved Stereo Matching Algorithm with Robustness to Noise Based on Adaptive Support Weight

“…The obtained 3D information can be used in a number of applications, such as automotive vehicles, robot vision, monitoring systems, mobile applications, and 3D reconstruction [1][2][3]. Crucial to the reliability of 3D information in these applications is the accurate calculation of disparity that is carried out by two types of stereo matching algorithms: local and global matching algorithms [4,5]. Global matching algorithms, although relatively accurate, require time-consuming and complex calculations while local matching algorithms, which have relatively low matching accuracy, entail fast and simple calculations.…”

Stereo matching for large-scale high-resolution satellite images using new tiling technique

“…Although feature-based methods have achieved a lot in the field of computer vision in recent years, most contemporary applications, especially remote sensing and photogrammetry, require dense disparity information which is attained by region-based matching methods [5] [6]. According to the recent survey [5], the region-based method employs scan-line algorithms, dynamic programming algorithms, graph-cut algorithms, and belief propagation algorithms, and there is also a need for dense disparity information.…”

“…According to the recent survey [5], the region-based method employs scan-line algorithms, dynamic programming algorithms, graph-cut algorithms, and belief propagation algorithms, and there is also a need for dense disparity information. Sub-regioning method [7] produces a dense disparity map by using rectangular sub-regioning (RSR) and two-stage have also studied this algorithm for much greater benefit.…”