Stereo matching, as many problems in computer vision, has been addressed by a multitude of algorithms, each with its own strengths and weaknesses. Instead of following the conventional approach and trying to tune or enhance one of the algorithms so that it dominates the competition, we resign to the idea that a truly optimal algorithm may not be discovered soon and take a different approach. We present a novel methodology for combining a large number of heterogeneous algorithms that is able to clearly surpass the accuracy of the most accurate algorithms in the set. At the core of our approach is the design of an ensemble classifier trained to decide whether a particular stereo matcher is correct on a certain pixel. In addition to features describing the pixel, our feature vector encodes the agreement and disagreement between the matcher under consideration and all other matchers. This formulation leads to high accuracy in disparity estimation on the KITTI stereo benchmark.
Aristotle Spyropoulos And Philippos Mordohai,
Stevens Institue of Technology, New Jersey, USA
PDF
Poster
The objective of our research is to test the hypothesis that a competitive stereo matching system can be constructed by combining multiple stereo matching algorithms in a principled way without requiring domain expertise from the users.
We designed a classifier ensemble that select among multiple candidate disparities for each pixel using a technique to select which of the over 100 available matchers to use as inputs to the classifiers.
SAD, SSD, Sobel, XNCC, SNCC, Censues, Shiftable Windows, MRF, rSGM, FCVF, ELAS, DAISY, Superpixels.
Distance from Disccontinuity, Left-Right Consistency.
Each figure highlights the pixels that selected the corresponding disparity of each matcher.
The top four (out of eight) matchers are shown. Click on each image for larger view.
inproceedings {spyr15_3dv,
title = {Ensemble Classifier for Combining Stereo Matching Algorithms},
author = {Spyropoulos, Aristotle and Mordohai, Philippos},
booktitle = {International Conference on 3D Vision (3DV)},
year = {2015}
}