Background information
Civil Maps enables on-demand perception and actuation of the world around us, by collecting and analyzing spatial data.
Tentative plan
- The evolution of surveying (in history, and in recent times)
- Lidar and photogrammetry technologies: point clouds, maps and reports
- The data problem (traditional approaches)
- Civil Maps approach: AI / Machine Vision solutions
- CM system architecture
- CM technical metrics (performance)
- Vision and future directions
Notes
Basic idea CivilMaps generates 3D instead of 2D for surveying / planning. They can do more, faster, for cheaper.
History of Surveying goes all the way back to primitive sites like Stonehenge.
Sensors
- LIDAR laser equivalent of RADAR, yields a point cloud
- Camera gets RGB value for each point from (close to) the same point of view
- Or, you can do multi-spectral LIDAR
How is this different than, e.g., Google street view?
This is a true 3D model constructed from points in a 3D space. Google street view is a 2D representation that allows us to get some 3D.
Can clearly identify individual objects in space. Application: figure out how high your train could be on a given stretch of track.
What are the limitations of the current system?
The browser can only handle so much data.
General Challenges
Standard “big data” story - can acquire data much faster than we can process it. Manual labelling of assets is very cumbersome.
How to quantify error? Points allow for a straightforward distance. Given the lack of standard, report a transparently computed measure of difference between manual and automatic labels.
Demo
- You can upload your own data (for now) here.
- Can choose features from a menu.
- Technology can work with a ~$500 LIDAR (as opposed to ~$100,000 survey quality LIDAR)
Visualizer Links
Barriers to humanitarian or conservation work
- Once spatial discriminations are complex (broken buildings, different vegetation), the AI isn’t there yet, but this should be possible.
- Identifying obstacles in roads could be a good application.