LiDAR Point Cloud Vectorization

In this project, I’m working on vectorizing LiDAR data using the Vancouver dataset. The goal is to take raw 3D point cloud data, classify it, and convert it into useful vector shapes, such as building footprints and vegetation outlines.

Classes

The LiDAR data includes the following classes:

• 0: Bare-earth/low grass
• 1: Low vegetation (<2m)
• 2: High vegetation (>2m)
• 3: Water
• 4: Buildings
• 5: Other
• 6: Noise (outliers, errors)

Overview

The process involves taking LiDAR data and identifying key elements like buildings, water bodies, and vegetation. After classifying the data, I vectorize it into 2D and 3D models. The project also focuses on filtering noise to improve the final output.

Dataset

I’m using the Vancouver dataset, which provides rich LiDAR data for experimentation. It’s already classified into different features like ground, buildings, and vegetation, making it ideal for vectorization.

Key Steps

1. Preprocessing: Set up the environment, load LiDAR data, and filter the points (e.g., removing noise).
2. Vectorization: Use the classified points to create 2D/3D models of buildings and other elements.
3. Automation: Automate the pipeline to process large datasets efficiently.

Goals

• Classify and vectorize key features in the point cloud.
• Generate clean 3D models from the data.
• Filter out noise and improve data quality for better results.

Running the Pipeline

1. Load the LiDAR dataset.
2. Classify the points.
3. Vectorize the results.
4. Visualize the 3D models with Open3D or export the results for further analysis.

Note: Open3D requires Python 3.11 or lower.

References

• Used this video for reference: YouTube Video