Integration of Absolute Orientation Measurements in the KinectFusion Reconstruction Pipeline

CVPR'18 Workshop on Visual Odometry and Computer Vision Applications Based on Location Clues

Available at openaccess.thecvf.com

@InProceedings{Giancola_2018_CVPR_Workshops,
  author = {Giancola, Silvio and Schneider, Jens and Wonka, Peter and Ghanem, Bernard S.},
  title = {Integration of Absolute Orientation Measurements in the KinectFusion Reconstruction Pipeline},
  booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
  month = {June},
  year = {2018}
}

Implementation

Tested with:

• VTK 6.2,
• Boost 1.58,
• Eigen 3,
• FLANN,
• Qt 5.9

Install a modified version of PCL

mkdir pcl-build && cd pcl-build/
cmake -DBUILD_GPU=ON -DBUILD_CUDA=ON -DWITH_QT=OFF ../pcl
make -j8
sudo make install
cd ..

Install a modified version of PCViewer

mkdir PCViewer-build && cd PCViewer-build
cmake ../PCViewer
make -j8 testKinFuGUI
cd ..

Create dataset

• Download the sequences from the freiburg dataset
• Extract the sequences into the data folder
• Activate the conda environment conda env create -f environment.yml && conda activate TUMdataset
• Copy python script into sequence folder cp *.py {sequencename} && cd {sequencename} with {sequencename} being the name of the folder for a given sequence
• Extract the orientated point cloud mkdir pointclouds && python generate_registered_pointcloud_organized.py rgb.txt depth.txt groundtruth.txt pointclouds/PC --pcd_format cd ../..

Run Kinect Fusion

cd PCViewer
./testKinFuGUI