Road_lanes_classification

Road_lanes_classification_from_pointclouds_using_machine_learning

Prerequisites

• Python version 3.10
• jupter notebook note: To ensure seamless integration, please modify the paths as per your specific dataset storage location

Required libraries

• pip install pandas
• pip install numpy
• pip install python-time
• pip install os-sys
• pip install scikit-learn
• pip install matplotlib
• pip install tensorflow
• pip install torch
• pip install torchvision
• pip install torchstat
• pip install Pillow
• pip install opencv-python
• pip install tqdm
• pip install seaborn

Approaches used

Approach 1 decision tree Approach 2 pointcloud2image Approach 3 point net Note: Approach 2 was employed due to the nature of the dataset, which consists of road lane segments. The primary focus lies in extracting valuable insights from the surface characteristics.

Dataset

Road lane segments as .npy

distributed into 6 clasess

1. 2lanes
2. 3lanes
3. crossing
4. split4lanes
5. split6lanes
6. transition

with 22 features

1. 0 local x
2. 1 local y
3. 2 local z
4. 3 red values
5. 4 green values
6. 5 blue values
7. 6 global x
8. 7 global y
9. 8 global z
10. 9 intensity
11. 10 number of lidar returns
12. 11 planarity
13. 12 linearity
14. 13 sphericity
15. 14 verticality
16. 15 mean intensity in 0.3m increments along y
17. 16 mean intensity in 1.5m increments along y
18. 17 mean intensity in 0.3m increments along x
19. 18 mean intensity in 1.5m increments along x
20. 19 edge area
21. 20 grid increment index 0.3m resolution
22. 21 intensity principal gradient positions

Note: I provided one example of the dataset