Reimplementation of the Probabilistic Occupancy Map (POM) for multi-view multi-person detection in C with Python wrapper.
F. Fleuret, J. Berclaz, R. Lengagne and P. Fua, Multi-Camera People Tracking with a Probabilistic Occupancy Map, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 30, Nr. 2, pp. 267 - 282, February 2008.
https://www.epfl.ch/labs/cvlab/software/tracking-and-modelling-people/pom/
- numpy
- scipy
- imageio
- pyyaml
- matplotlib
- python 3.5+
- CMake (c++ only if built manually)
- Boost.Python 1.66+
- Follow the instruction on how to install Boost.Python in the section below
- Compile the core using
build.sh- In case of error try to compile the core manually. Have a look at
build_manually_cmake.shorbuild_manually_macos.sh. You will have to find the location of the headers and libraries of python, boost-python and boost-numpy and modify paths in these files. Trysudo find / -name "libboost_python*"if don't know where to look.
- In case of error try to compile the core manually. Have a look at
- If the core is compiled correctly you should be able to run this
cd pypom; python -c "import core; help(core)" - Add the package in your environmental variable
export PYTHONPATH="~/code/pypom:$PYTHONPATH" - That's it! You should be able to import pypom in python.
We strongly suggest installing Boost.Python from source as the headers files are required to compile our code:
cd $HOME
wget https://dl.bintray.com/boostorg/release/1.66.0/source/boost_1_66_0.tar.bz2
tar --bzip2 -xf boost_1_66_0.tar.bz2
cd boost_1_66_0
./bootstrap.sh --with-python=python3
sudo ./b2 install
sudo ldconfig
- If you have a message similar to this one:
failed updating 66 targets...fatal error: pyconfig.h: No such file or directorythen you need to modify line 542 of file tools/build/src/tools/python.jam from this: includes ?=$(prefix)/include/python$ (version) ; to this: includes ?=$(prefix)/include/python$ (version)m ; now try again installing boost.python:
./bootstrap.sh --with-python=python3
sudo ./b2 install
sudo ldconfig
- Another version of Boost is clashing with the new one. If it is the case for you we suggest removing the old version and re-install the new one. HINT: after compiling the core module, type
ldd core.soto see which boost.python was used.
On MacOS Boost.Python can be installed with Homebrew:
brew install boost
brew install boost-python3
brew install cmake
take note of which python version it has been used to install boost-python! If you have multiple python installed or Anaconda you have to make sure, while compiling this project, that the correct versions are used.
If there is a mismatch of version try to build this project manaully. Have a look at the files build_manually_cmake.sh or build_manually_macos.sh.
If you have no clue where to look:
sudo find / -name "libboost_python*"
sudo find / -name "libboost_numpy*"
sudo find / -name "libboost*"
GPL v3
- placing the cameras at an altitude is better than close to the floor
- the more the cameras the more the people you can detect correctly
- place the cameras so as the picture is straight otherwise the rectangles will be larger and performance will degrade
- obstacles such as walls or forniture are not handeled by the algorithm. This means that it will fail in such a scenarios
- using a CNN to perform background subtraction is in general the best option
- shadows are really bad. Try to avoid them as much as possible
- make sure the cameras are well calibrated and synchronized
The installation of CMake with SSL support is required for the installation of OpenCV with the special Intel IPP functions.
sudo apt-get install libcurl4-gnutls-dev
mdkir -p ~/tmp/cmake
cd !$
wget https://cmake.org/files/v3.10/cmake-3.10.2.tar.gz
tar -xzf cmake-3.10.2.tar.gz
cd cmake-3.10.2/
./bootstrap --parallel=8 --system-curl
make -j 8
then we remove the old CMake with:
sudo apt remove cmake
and finally we install the new CMake with:
sudo make install