LIBELAS: Library for Efficient Large-scale Stereo Matching
LIBELAS (Library for Efficient Large-scale Stereo Matching) is a cross-platfrom (Linux, Windows) C++ library with MATLAB wrappers for computing disparity maps from rectified graylevel stereo pairs. It is robust against moderate changes in illumination and well suited for robotics applications with high resolution images. Computing the left and right disparity map of a one Megapixel image requires less than one second on a single i7 CPU core. More details can be found in our related publications.
30.06.2015: Fixed a little bug in the OpenMP version, thanks to Giulia Pasquale for reporting.
03.02.2015: Added some header includes for compatibility with Visual Studio 2013, thanks to Xiaoyan Hu for reporting.
15.10.2014: Fixed a bug when extrapolating to the image boundaries, thanks to Daniel Scharstein for reporting.
05.09.2014: Fixed a bug in the maximum computation, thanks to Jun Fu for reporting.
21.11.2013: Changed the flags passed to the compiler such that the library builds on Mac OS X 10.9 again. Thanks to Winston Churchill and Alastair Harrison for reporting this issue.
07.11.2013: Fixed a bug in the triangulation code that occurs on 64-bit systems with more than 4GB memory, thanks to Marko Markovic for reporting.
14.03.2012: Fixed another bug in adaptive smoothing, removed a momory leak.
19.09.2011: Fixed the shifting bug of adaptive smoothing, a small bug when finding support points as well as a couple of memory leaks: Thanks to Simon Hermann, Markus Moll and Wojciech Chojnowski for reporting those issues.
26.05.2011: Fixed a flipped parameter bug in the Matlab demo code, thanks to Diego Cheda for reporting
24.01.2011: New version uploaded! New features: SSE instructions used for computing descriptor (faster), better approximation to bilateral filtering for postprocessing, bytes-per-line parameter for input images. Further, a new parameter 'subsampling' allows for evaluating only every second pixel, which is often sufficient in robotics applications, since depth accuracy matters more than a large image domain. Using this option, we run ELAS at 8-10 fps on the 0.6 Megapixels Karlsruhe data set using a single i7 3Ghz core.
12.11.2010: Minor changes for running on Mac, thanks to Tobias Feldmann
11.11.2010: Fixed a bug in Elas::createGrid, thanks to Samuele Salti for reporting
03.10.2010: Added support for minimum disparity parameter (must be >=0)
20.09.2010: First version online
We tried to keep dependencies as small as possible, but to get started you are going to need at least: