LOAM and Mask R-CNN Semantic Mapping: semantic labelling for lidar point cloud with a color and a depth camera. Tested with Velodyne VLP-16, Intel D435i, Ubuntu 20.04, OpenCV 4.5.4 and ROS noetic.
LOAM: https://github.com/laboshinl/loam_velodyne/
Mask R-CNN: https://github.com/matterport/Mask_RCNN/
Mask R-CNN model example: https://github.com/opencv/opencv/wiki/TensorFlow-Object-Detection-API
Color file and Mask R-CNN class name file are provided in ./mask_rcnn_documents
Examples of calibration data are provided in ./calib_params
Before running, please change the path of intrinsic and extrinsic parameter files, color file, Mask R-CNN class name, model and config files in ./src/lib/BasicSemanticMapping.cpp
Sample map, opened with rviz.
Before making, it is needed to adapt cv_bridge package of ROS to use OpenCV built with GPU support.
$ cd ${any path you like}
$ mkdir -p catkin_ws/src
$ cd ./catkin_ws/src
$ git clone https://github.com/Yinshideguanghui/loam_maskrcnn_semantic_mapping.git
$ catkin_init_workspace
$ cd ..
$ catkin_make -DCMAKE_BUILD_TYPE=Release
$ source ./devel/setup.bash
roslaunch loam_velodyne_semantic_mapping loam_semantic_mapping.launch
Issues #71 and
#7 address this
problem. The current known solution is to build the same version of PCL that
you have on your system from source, and set the CMAKE_PREFIX_PATH
accordingly so that catkin can find it. See this
issue
for more details.
Quantifying Aerial LiDAR Accuracy of LOAM for Civil Engineering Applications. Derek Anthony Wolfe