virtualroyalty/pointcloudsegmentation

The research project based on Semantic KITTTI dataset, 3d Point Cloud Segmentation , Obstacle Detection

PointCloudSegmentation

---

---

---

Project structure:

├───docker-env/
├───obstacle-detection/
│   ├───dataset/
│   │   └───sequences/
│   │       └───00/
│   │           ├───clusters/
│   │           ├───labels/
│   │           └───velodyne/
|   ├───model/
|   |
│   ├───examples/
│   │   
│   ├───pipeline/
│   │  
│   └───scripts/
│       
└───visualization/

How to dockerize this:

---

• In base-notebook/ folder start Docker and build an image: $ docker build -t jupyter .
• After that you can verify a successful build by running: $ docker images
• Then start container by running:
  
  $ docker run -it --rm -p 8888:8888 -v /path/to/obstacle-detection:/home/jovyan/work jupyter
  
  NOTE: on Windows you need to convert your path into a quasi-Linux format (e.g. //c/path/to/obstacle-detection). More details here
  Also, if you want to use drive D:/ you need to check whether it is mounted or not and if not mount it manually. More details here if you use Docker toolbox
  
  
• After correct running you will see URL to access jupyter, e.g.:
  
  httр://127.0.0.1:8888?token=0cccd15e74216ed2dbe681738ed0f9c78bf65515e94f27a8
  
  
• To access jupyter you need to go for Docker IP:8888?token=xxxx...
  ( e.g. httр://192.168.99.100:8888/?token=0cccd15e74216ed2dbe681738ed0f9c78bf65515e94f27a8)
  
  
• To enter a docker container run $ docker exec -it *CONTAINER ID* bash (find out ID by running $ docker ps)

Pre-trained Models

SemanticKITTI

• squeezeseg
• squeezeseg + crf
• squeezesegV2
• squeezesegV2 + crf
• darknet21
• darknet53
• darknet53-1024
• darknet53-512

References and useful links:

---

Dataset:

1. Web-site Semantic KITTI
2. Paper Semantic KITTI

Segmentation:

3. Segmentation approaches Point Clouds
4. Also about point cloud segmentation
5. PointNet
6. PointNet++ from Stanford
7. PointNet++
8. RangeNet++

Obstacle detection:

9. Obstacle Detection and Avoidance System for Drones
10. 3D Lidar-based Static and Moving Obstacle Detection
11. USER-TRAINABLE OBJECT RECOGNITION SYSTEMS
12. Real-Time Plane Segmentation and Obstacle Detection

Useful Github links:

13. https://github.com/PRBonn/semantic-kitti-api
14. https://github.com/jbehley/point_labeler
15. https://github.com/daavoo/pyntcloud
16. https://github.com/strawlab/python-pcl
17. https://github.com/kuixu/kitti_object_vis
18. https://github.com/lilyhappily/SFND-P1-Lidar-Obstacle-Detection
19. https://github.com/kcg2015/lidar_ground_plane_and_obstacles_detections
20. https://github.com/enginBozkurt/LidarObstacleDetection