Rail Scene Lidar Classification

Job ID: 39342648

Budget: €30 – €250 EUR

I need lidar data from a rail scene classified. The data includes:

- Rail tracks and switches
- Electrical and signaling equipment
- Station platforms and signage
- Poles, platforms, crossings, and overhead structures
LiDAR Point Cloud Semantic Segmentation for Railway Infrastructure

I am seeking an experienced freelance data scientist or machine learning engineer to work on semantic segmentation of LiDAR point clouds, specifically for railway infrastructure detection. The objective is to accurately classify key railway components such as rail tracks, overhead wires, poles, platforms, and other structural elements from raw LiDAR (.laz).

The model should work dynamically on any terrain and on any CRS (coordinates) file type.
I am expecting similar like "https://github.com/akharroubi/Rail3D"


Project Scope:

Develop and optimize point cloud preprocessing pipelines, including normalization, downsampling, and augmentation strategies.

Implement advanced deep learning models (any best-suited model) tailored for point cloud semantic segmentation.

Address class imbalance and ensure robust detection across diverse geographical regions.

Conduct model validation, performance tuning, and detailed analysis of segmentation results.

Provide visualization of classified point clouds with intuitive color-coding for clearly identifiable railway components.

Technical Requirements:

Proficiency with Python, PyTorch, or TensorFlow

Experience in handling large-scale LiDAR datasets

Familiarity with libraries such as PDAL, Open3D, LasPy, and visualization tools like Potree

Understanding of geospatial coordinate systems and point cloud normalization techniques

Deliverables:

Optimized segmentation model achieving high accuracy (F1, mIoU metrics)

Efficient, scalable processing pipeline

end to end code pipeline