Depth packet processors and plot functions for Kinect v2 phase unwrapping from log files.
The algorithm is described in the paper Efficient Phase Unwrapping using Kernel Density Estimation, ECCV 2016, Felix Järemo Lawin, Per-Erik Forssén and Hannes Ovrén, see http://www.cvl.isy.liu.se/research/datasets/kinect2-dataset/.
git clone https://github.com/felja633/kinectv2_decoders.git
Download dataset at www.cvl.isy.liu.se/research/datasets/kinect2-dataset/kinect2_dataset.zip.
Unzip kinect2_dataset.zip in the kinectv2_decoders folder
then build the decoders:
cd kinectv2_decoders
mkdir build
cd build
cmake ..
make
We provide three datasets: lecture, kitchen, and library. Choose one as dataset and run the code as:
cd kinectv2_decoders/build
./kinectv2_decoders ../parameters/default_setup.xml dataset
cd ..
python evaluate_decoders.py test parameters/default_setup.xml dataset
To visualize, run code as:
python evaluate_decoders.py vis ../parameters/default_setup.xml dataset
Toggle between frames using the arrow buttons.
Parameters are passed in xml-format. At this stage two pipelines are implemented, kde and libfreenect2. Each pipeline that is to be tested should be added in the xml-file. The user can then add and change the parameters freely.
Example:
<pipeline name="kde" setup_name="base">
<Parameters>
<kde_sigma_sqr>0.0239282226563</kde_sigma_sqr>
<unwrapping_likelihood_scale>2.0</unwrapping_likelihood_scale>
<phase_confidence_scale>3.0</phase_confidence_scale>
<kde_neigborhood_size>5</kde_neigborhood_size>
<num_hyps>2</num_hyps>
<min_depth>500.0</min_depth>
<max_depth>18750.0</max_depth>
</Parameters>
</pipeline>The package requires the hdf5 library to parse Kinect v2 log files.
First Install HDF5 with brew:
brew install homebrew/science/hdf5
Then add the appropriate paths (in bash):
export HDF5=/usr/local/Cellar/hdf5/1.8.17 # Replace with your hdf5 installation full path
export PATH=${HDF5}/bin:${PATH}
export DYLD_LIBRARY_PATH=${HDF5}/lib:${DYLD_LIBRARY_PATH}
sudo apt-get install libhdf5-dev