Implementation of A Random Finite Set Approach for Dynamic Occupancy Grid Maps with Real-Time Application

This repository is fast moving and we currently guarentee no backwards compatibility.

* increasing the grid cell count to 1.44 * 10⁶ increases the runtime by only ~20ms

** the NVIDIA GTX 980 is more than twice as fast as the NVIDIA GTX 1050 Mobile

You need OpenCV, OpenGL, GLFW3, GLEW, GLM and CUDA to compile and run this project. You can find the setup instructions for Ubuntu (tested for 18.04 LTS and 20.04 LTS) and Windows 10 below.

• OpenCV:

  sudo apt install libopencv-dev

• GLFW3, GLEW, GLM:

  sudo apt install libglfw3-dev libglew-dev libglm-dev

• CUDA:

  • Have the most recent nvidia driver installed on your system, then check which CUDA version it supports

    nvidia-smi -q | grep CUDA

  • Follow the corresponding instructions from the cuda toolkit archive

  • Update environment variables in your bashrc (or equivalent rc file)

    echo "\n# CUDA paths\nexport PATH=$PATH:/usr/local/cuda/bin\nexport CUDADIR=/usr/local/cuda\nexport LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64\n" >> ~/.bashrc

  • Reboot (if necessary, a couple of times)

  • Check that your system GPU driver is still working by executing nvidia-smi

  • To verify that your CUDA installation is working, consider compiling and running one of the examples from /usr/local/cuda/samples. In particular ones that use OpenGL+CUDA, e.g. samples/2_Graphics/Mandelbrot.

• OpenCV:

  • Download and install OpenCV Windows from here.

• GLFW3, GLEW, GLM:

  • It is recommended to place all packages in a folder dogm/external

    • Download GLFW3 64-bit Windows binary, GLEW binaries and GLM
    • Unzip all of them in the dogm/external folder.

  • In the end you should have a folder structure that's something like this:

    • dogm/external/glew-2.1.0
    • dogm/external/glfw-3.3
    • dogm/external/glm

• CUDA:

  • Have the most recent nvidia driver installed on your system, then check which CUDA version it supports
  • Follow the corresponding instructions from the cuda toolkit archive

Use CMake to build the project:

Use the local pipeline to configure, build and execute the project:

./local_pipeline_ubuntu.sh

The pipeline creates folder build and compiles executables into that folder. Call the pipeline with -h to get an overview of optional flags.

On Windows it's easiest to use cmake-gui. Use it to configure and generate the project like shown below (required variables are marked with a red dot):

Afterwards open the generated .sln in Visual Studio 17/19 and compile it.

Contributions are welcome. Please make sure to apply clang-format to your code, e.g. by letting local_pipeline_ubuntu.sh check and fix formatting for you. We also recommend to include formatting with clang-format in your editor/IDE.

Nuss et al. "A Random Finite Set Approach for Dynamic Occupancy Grid Maps with Real-Time Application"

Dominik Nuß. "A Random Finite Set Approach for Dynamic Occupancy Grid Maps"

Homm et al. "Efficient Occupancy Grid Computation on the GPU with Lidar and Radar for Road Boundary Detection"

mitkina/dogma has also been a great reference whenever I got stuck.