/Lidar-Road-Marking-Segmentation

Extracting Road Markings Using Only Lidar PointCloud

Primary LanguagePythonApache License 2.0Apache-2.0

Lidar Road Marking Segmentation

This repo is based on cylinderical 3d paper and code, this repo is actually a fork of cylinderical 3d repo

This code is tested on Ubuntu 20.04 and CUDA version 11.4

To run the code on newer CUDA version, you may need to check the new updates in the original repo of cylinderical 3d model(the one we forked from)

we trained the cylindircal 3d model on nuscenes, intially it was trained only on KITTI.

Our approach is tested on nuscenes dataset only as it is the one that have the drivable surface labels

Model weights link: https://drive.google.com/file/d/1mCXL2INlabwm7ExOvhgponQUBQNM6HKf/view?usp=sharing

Medium Article Explaining the pipeline: https://mohamedelbassiouny.medium.com/lidar-road-marking-segmentation-d4d7cd437808

Data in folders

  • lidar_data/ put in it the raw lidar bins of nuscenes having structure of (x,y,z,intensity,ring), so each bin have size of N x 5, where N is the number of points in the lidar scan.
  • model_load_dir_nuscenes/ put in the weights of the trained model, name must be model_weight.pt , you can change the path though from config/nuScenes.yaml
  • After inference, lidar_data_labels_all/ folder will have a label for each lidar bin, the label mapping can be found in config/label_mapping/nuscenes.yaml, the labels used in this yaml file are labels_16
  • After running the lane_marking_segmentation.py script, the lidar_data_labels_road_marking/ folder will have label files also, but with label 1 for road marking and 0 for non-road marking.

How to run the code

  • install requirements conda create --name <env> --file environment.yml
  • conda activate <env>
  • run python inference_nuscenes.py to invoke cylinderical 3d model, labels will be saved in lidar_data_labels_all/ folder
  • run python landmarks.py, pointcloud and its labels will be saved in lidar_data_labels_road_marking/ and visualisation images will be saved in output_vis_folder/

End result (red is road marking)

Alt Text