Lidar Point–to–point Correspondences for Rigorous Registration of Kinematic Scanning in Dynamic Networks

This repository contain an application written in C++ that integrates point-to-point correspondences between LiDAR point clouds with raw inertial and GNSS measurements using Dynamic Networks. The provided code is based on the ROAMFREE sensor fusion library, which contains the actual solver for Dynamic Network adjustment problems.

The algorithms and methods are presented in detail in the following article:

• Brun, Aurélien, Davide Antonio Cucci, and Jan Skaloud. "LiDAR Point--to--point Correspondences for Rigorous Registration of Kinematic Scanning in Dynamic Networks." arXiv preprint arXiv:2201.00596 (2022). (link)

The code provided in this repository allows to obtain Trajectory 2 DNC as presented in Section 3.3: nominal (optimal GNSS reception) of the article above. With minor modifications either in the code or in the measurement data, all DN* trajectories can be obtained. For example, to obtain the estimated trajectories with simulated GNSS outages, T.8 DNCo and T.9 DNCo1, it is sufficient to remove records from the GNSS measurement file in the data folder.

Obtain the code

This repository can be cloned and the submodules initialized with the following commands:

git clone https://github.com/TOPO-EPFL/DN-LiDAR.git
git submodule update --init

Build

ROAMFREE relies on Eigen3, and on a modified version of g²o, which in turn relies on suitesparse. Few functions from opencv are also used. A compiler that supports C++11 is required.

On Ubuntu (e.g., 20.04) all required dependencies can be installed with the following command:

sudo apt-get install build-essential cmake \
                     libsuitesparse-dev libeigen3-dev libboost-all-dev \
                     libopencv-dev

Then, the code can be built from the repository root folder as follows:

mkdir build
cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
make

Run

Once the code has been built, an executable named dn will be available in the build folder. If run, it will provide the following output

$ ./dn 
 * Configuring solver ... 
 * Adding measurements ... 
 * Solving ...  done, final chi2 = 206511.090
 * Compressing results in ../data/trajectories/T2-DNC.tar.gz
 * Removing temporaries

An archive named T2-DNC.tar.gz is generated in the data/trajectory folder of the repository. This archive contains the trajectory referred as T.2 DNC in the referenced paper.

Visualizing the output

A matlab script is provided to compare the estimated trajectory, T.2 DNC and the one obtained with Kalman smoothing, T.1 KS with the ground-truth one, T.0 R, as presented in Section 3.3: nominal (optimal GNSS reception) of the article above.

Running the script tools/trajectory_comparison/runTrajectoryComparison.m will produce, among the others, the following two plots depicting the orientation error with respect to the ground truth.