This is a port of the code for the paper entitled Foreground detection for moving cameras with stochastic approximation, adapted to work in Python 2.7 (some late version, such as 2.7.15). The original code is copyrighted by Francisco Javier Lopez Rubio. All modifications are copyrighted by Jose David Fernandez Rodriguez.
It is easy to port the C++ code to use newer versions of OpenCV. However, most Debian-based distros do not come with official, off-the-shelf packages for OpenCV-contrib's nonfree module, so in most cases you'd need either of:
- Download a binary package from an untrusted source
- Compile OpenCV and OpenCV-contrib from scratch
- Compile just the nonfree module, or just the SURF sources.
And the C++ sources might have to be modified depending on what version of OpenCV is available!
For now, this version works just fine if you compile OpenCV-2.4.8 in the parent folder (i.e. ..; it
is enough to compile the shared libraries).
The original code also uses Intel's TBB, using all of its three libraries (tbb, tbbmalloc and tbbmalloc_proxy).
Unfortunately, there seem to be some incompatibilities that prevent the use of tbb_malloc_proxy. This might be
solvable, but does not warrant the effort right now. The easiest option, and the one chosen here, is to
disable the use of the latter library by commenting out all instances of #include <tbb/tbbmalloc_proxy.hpp>.
The original code works with some older version of Matlab (like R2014b or R2015b, not sure about that). This repo also has a branch (matlabr2017b) with modifications to the original code to make it work in Matlab R2017b. Because of incompatibilities with Matlab R2017b's TBB version, this branch has tbbmalloc_proxy disabled.
Example data not included, but can be found at Ezequiel's research webpage.
Dependencies:
- cython 0.28.2
- numpy 1.14.3
- scipy 1.1.0
- imageio 2.3.0
- matplotlib 2.2.2
- OpenCV 2.4.8
- Intel's TBB
- Eigen, the C++ matrix header-only library