Efficient Sampling of Disparity Space for Fast And Accurate Matching

Abstract

A simple stereo matching algorithm is proposed that visits only a small fraction of disparity space in order to find a semi-dense disparity map. It works by growing from a small set of correspondence seeds. Unlike in known seedgrowing algorithms, it guarantees matching accuracy and correctness, even in the presence of repetitive patterns. This success is based on the fact it solves a global optimization task. The algorithm can recover from wrong initial seeds to the extent they can even be random. The quality of correspondence seeds influences computing time, not the quality of the final disparity map. We show that the proposed algorithm achieves similar results as an exhaustive disparity space search but it is two orders of magnitude faster. This is very unlike the existing growing algorithms which are fast but erroneous. Accurate matching on 2-megapixel images of complex scenes is routinely obtained in a few seconds on a common PC from a small number of seeds, without limiting the disparity search range.1 The term disparity component has been coined in [1].

Joint disparity and optical flow by correspondence growing

“…We compared the proposed algorithm which jointly estimates disparity and optical flow (scene-flow, green circles) with other seed growing algorithm which separately computes the disparity and optical flow frame-by-frame independently [11,10] (stereo resp. flow, red crosses).…”

“…A basic principle of the seed growing methods is that correspondences are found in a small neighborhood around given initial seed correspondences. This idea has been adopted in stereo [10,11,12,13]. The advantage of such approaches is a fast performance compared to global MRF methods, and a good accuracy compared to purely local method, since neighboring pixel relations are not ignored completely.…”

“…Notice that number of matched pixels in D 1 is always smaller or equal than in D 0 , since the candidates in the neighborhood (3) do not exist in locations where D 0 is undefined. Therefore, when next frame comes, disparity map D 0 := D 1 and this disparity map is further grown again by the left-right stereo algorithm [11], to recover motion occluded pixels. Without this recovery, the algorithm would in fact track the initial points in the first disparity map over time, but it will sooner or later end up with an empty result.…”

“…First disparity map D 0 is obtained by (also seed growing) stereo algorithm [11]. Notice that number of matched pixels in D 1 is always smaller or equal than in D 0 , since the candidates in the neighborhood (3) do not exist in locations where D 0 is undefined.…”

High-resolution depth maps based on TOF-stereo fusion

“…A practical tradeoff between the local and the global methods in stereo is the seed growing class of algorithms [5], [6], [4]. The correspondences are grown from a small set of initial correspondence seeds.…”