NemoLeiYANG / InfrasCal

InfrasCal: Automatic Infrastructure-based Intrinsic and Extrinsic Calibration of a Multi-camera System

InfrasCal

Introduction

This C++ library supports the following tasks:

1. Extrinsic infrastructure-based calibration of a multi-camera rig
2. Intrinsic and extrinsic infrastructure-based calibration of a multi-camera rig

The following two camera models are supported in this library:

• Pinhole camera model with radial and tangential distortion
• Equidistant fish-eye model (J. Kannala, and S. Brandt, A Generic Camera Model and Calibration Method for Conventional, Wide-Angle, and Fish-Eye Lenses, PAMI 2006)

The infrastructure-based calibration runs in near real-time, and is strongly recommended if you are calibrating multiple rigs with mapping datasets.

The workings of the library are described in the three papers:

    Yukai Lin, Viktor Larsson, Marcel Geppert, Zuzana Kukelova, Marc Pollefeys, Torsten Sattler,
    Infrastructure-based Multi-Camera Calibration using Radial Projections, accepted on ECCV2020.

    Lionel Heng, Mathias Bürki, Gim Hee Lee, Paul Furgale, Roland Siegwart, and Marc Pollefeys,
    Infrastructure-Based Calibration of a Multi-Camera Rig,
    In Proc. IEEE International Conference on Robotics and Automation (ICRA), 2014.
    
    Lionel Heng, Paul Furgale, and Marc Pollefeys,
    Leveraging Image-based Localization for Infrastructure-based Calibration of a Multi-camera Rig,
    Journal of Field Robotics (JFR), 2015.

If you use this library in an academic publication, please cite at least one of the following papers depending on what you use the library for.

Acknowledgements

The InfrasCal library includes third-party code from the following sources:

    1. Lionel Heng, Bo Li, and Marc Pollefeys,
       CamOdoCal: Automatic Intrinsic and Extrinsic Calibration of a Rig
       with Multiple Generic Cameras and Odometry,
       https://github.com/hengli/camodocal

    2. Sameer Agarwal, Keir Mierle, and Others,
       Ceres Solver.
       https://code.google.com/p/ceres-solver/
    
    3. D. Galvez-Lopez, and J. Tardos,
       Bags of Binary Words for Fast Place Recognition in Image Sequences,
       IEEE Transactions on Robotics, 28(5):1188-1197, October 2012.

    4. L. Kneip, D. Scaramuzza, and R. Siegwart,
       A Novel Parametrization of the Perspective-Three-Point Problem for a
       Direct Computation of Absolute Camera Position and Orientation,
       In Proc. IEEE Conference on Computer Vision and Pattern Recognition, 2011.

    5. pugixml
       http://pugixml.org/

    6. Changchang wu,
       SiftGPU: A GPU implementation of David Lowe's Scale Invariant Feature Transform
       http://cs.unc.edu/~ccwu
       
    7. Viktor Larsson, Torsten Sattler, Zuzana Kukelova and Marc Pollefeys,
       Revisiting Radial Distortion Absolute Pose.
       https://github.com/vlarsson/radialpose

Build Instructions for Ubuntu

Required dependencies

• BLAS (Ubuntu package: libblas-dev)
• Boost (Ubuntu package: libboost-all-dev)
• Eigen3 (Ubuntu package: libeigen3-dev)
• SuiteSparse (Ubuntu package: libsuitesparse-dev)
• Ceres-solver (Ubuntu package: libceres-dev)
• CUDA
• OpenCV+contrib

Optional dependencies

• GTest
• Glog (Ubuntu package: libgoogle-glog-dev)

Tested configuration versions

• Ubuntu 18.04
• Ceres 1.13.0
• Eigen 3.3.4
• OpenCV & opencv_contrib 3.4.2, 4.1.1
• Boost 1.65.1
• Cuda 9.1, 10.1

1. Before you compile the repository code, you need to install the required dependencies, and install the optional dependencies if required.

   • Install Cuda

   • Build required dependencies

     sudo apt-get install cmake git gcc-6 g++-6 libopenblas-dev libblas-dev libeigen3-dev libgoogle-glog-dev 
     sudo apt-get install build-essential libgl1-mesa-dev libglu1-mesa-dev freeglut3-dev libglew-dev
     sudo apt-get install libatlas-base-dev libsuitesparse-dev libsqlite3-dev libceres-dev libboost-all-dev
     

   • Build Opencv

     mkdir -p ~/dev && cd ~/dev
     git clone --depth 1 --branch 3.4.2 https://github.com/opencv/opencv.git
     git clone --depth 1 --branch 3.4.2 https://github.com/opencv/opencv_contrib.git
     cd opencv && mkdir build && cd build
     CC=/usr/bin/gcc-6 CXX=/usr/bin/g++-6 cmake .. -DWITH_CUDA=ON -DCMAKE_BUILD_TYPE=Release \
     -DOPENCV_EXTRA_MODULES_PATH=../../opencv_contrib/modules -DOPENCV_ENABLE_NONFREE:BOOL=ON \
     -DCUDA_NVCC_FLAGS=--expt-relaxed-constexpr
     make -j8
     sudo make install
     

2. Build the code.

   mkdir -p ~/dev && cd ~/dev
   git clone https://github.com/youkely/InfrasCal.git 
   cd InfrasCal && mkdir build && cd build
   CC=/usr/bin/gcc-6 CXX=/usr/bin/g++-6 cmake -DCMAKE_BUILD_TYPE=Release ..
   make -j8
   

Examples

Go to the source folder. To see all allowed options for each executable, use the --help option which shows a description of all available options.

1. Infrastructure-based calibration

    ./build/bin/infrastr_calib --camera-count 5 \
    --output ./data/demo/results \
    --map ./data/demo/map \
    --database ./data/demo/map/database.db \
    --input ./data/demo/ \
    --vocab ./data/vocabulary/sift128.bin \
    -v --camera-model pinhole-radtan --save
   

   The camera-model parameter takes one of the following two values: pinhole-radtan, and pinhole-equi(kannala-brandt).

   The calibration mode takes one of the following options: InRaSU(default, corresponds to Inf+1DR+RA in the ECCV2020 paper), In(Inf+K), InRI(Inf+K+RI), InRa(Inf+RD), InRaS(Inf+RD+RA)