3D Lane Detection from LiDAR
3D lanes lane lines extracted directly from the LiDAR point cloud.
- Client
- Confidential
- Year
- 2024
- Category
- Computer Vision
- Role
- Design + Build
- Timeline
- 4–5 weeks
The problem
Camera lane detection fails in the dark, in glare and in bad weather, exactly when a vehicle needs it most. LiDAR keeps working, but its point cloud is sparse and noisy.
What we built
A two-stage pipeline: an intensity-threshold pass pulls the most likely lane-line points out of the point cloud, then a region-of-interest pass tightens the result to the drivable corridor. Output is 3D lane geometry, not a 2D image overlay.
The result
Lane geometry that holds up in conditions where a camera gives nothing, usable on its own or fused with vision.
- lane geometry in real-world coordinates
- 3Dlane geometry in real-world coordinates
- works in darkness, glare and weather
- no cameraworks in darkness, glare and weather
- intensity threshold → region of interest
- 2-stageintensity threshold → region of interest
Built with
- Python
- LiDAR
- Point cloud processing
- NumPy
- Open3D
We deliver what we commit.
Tell us what you're trying to build.
We'll come back within 24 hours with honest feedback on scope, timeline and cost, whether or not we turn out to be the right fit.