census test

Testing of Census-like stereo algorithms. Mostly a space for experimenting with small concepts and ideas. As such, the code has basically no comments and poor interfaces for most of it's functionality.

Algorithms Implemented

• SSD Block Matching (with an optional Sobel prefilter), for Left + Right frame perspective, with quadratic subpixel estimation
• Census Matching (24 bit, 7x7 descriptor) + Block Matching, for Left + Right frame perspective, with linear subpixel estimation
• Census (left frame only) with thresholds from the Intel RealSense R200. Implemented as documented in Intel's released documentation. Optionally includes a domain-transform on the cost volume. As published these thresholds can be effective in removing spurious matches.
• Semiglobal Matching (5 paths), with SAD + Census cost metrics, subpixel matching, R200 thresholds, discontinuity scaling for SGM, naive hole filling and bilateral filter window weights. Only on the left-frame perspective.

Dependencies

• C++14 compliant compiler
• GLFW3 for visualization
• librealsense for librealsense executable

Building

Windows

• Pull down a version of this repository
• A Visual Studio 2015 Solution is included in msvc/
• CensusMatching is the primary project of interest
• GLFW3 is installed via a NuGet Package, so Visual Studio should pull it down automatically

Linux

• Pull down a version of this repository
• Type make in the top-level-centest. This'll build everything and generate binaries in the Makefile directory.
• Makefile can be modified to build a debug version of the executable

Running

• Run via ./centest <json>
• An example would be ./centest github_census/github_census.json
• The example code expects to have a converted version of Middlebury's v3 dataset
• The conversion routine is on github_census/generate_middlebury.py. This is a python script that requires Pillow and Imagemagick.
• This program also supports passing valid json on the command line, instead of a file.
• Output results are in the current working directory, to an 32-bit float disparity image PFM, and a confidence map in 32-bit floating point.
• To visualize the pfm files, a vis_pfm script is provided ./vis_pfm depth.pfm [conf.pfm] with an optional conf.pfm argument, which generates an out.png file the execution directory with a histogram colored depthmap.

License

Mozilla Public License 2.0. More information is available on the Wikpedia article on MPL or MPL's official FAQ.