LIDAR data is really cool, but is it useful?
What it does
Renders LIDAR data and asset location data in a web browser. Let's users associate specific data points with an asset.
How we built it
Recipe for a new data labeling tool:
- Collect 360 Million LIDAR points into one point cloud.
- Use an octree based viewer for efficient rendering in the browser.
- Spin up a backend server, and populate it with a tonne of asset data.
- Teach the viewer to plot asset data.
- Teach the backend to store clipping data. (Backend and frontend are now friends, and will miss each other when apart)
Challenges we ran into
Network Rail is the champion of the acronym fan club! Figuring out what each file represented was a significant challenge, especially with no knowledge of the space.
Vendor lock in, sucks. The most useful, high resolution LIDAR data was stuck in a proprietary format so we had to rethink our entire approach and goal.
It's really hard to do classification/detection on point clouds, which would have been the most useful outcome.
Accomplishments that we're proud of
We made a really nice looking thing. It's super shiny.
We also think it's something that we would actually use if we wanted to work with this data more, and get precise point-wise labelled data.
What we learned
With efficient indexing any data is usable, even in a web browser.
Targets can change, and that's okay, the outcome can still be awesome.
LIDAR data is really interesting, and going to be a fast moving space in machine learning.