Optimized Routing with Satellite Imagery and Computer Vision
summary
Claims an end-to-end system called CRESI baseline that extracts road graphs plus per-segment speeds from satellite imagery and uses them for true time-optimal routing; it trains on SpaceNet5 (about 20,000 km labeled across cities) and generalizes to an unseen city in competition results. It runs fast — roughly 1 km² per minute on CPU — and produces routable networks with accurate travel-time estimates, aiming to replace slow manual efforts (e.g., 70+ days to map Puerto Rico).
AI-written summary · about
chapters (17)
- 0:00Intro
- 1:00Motivation
- 2:18Rogue Properties
- 3:25Data
- 4:18Challenges
- 4:52Space Knight 5
- 5:21Outputs
- 5:48Road Extraction
- 6:53Scale Extraction
- 7:16Example
- 9:01Road Tracer
- 9:30Road Tracer Speed
- 11:12Travel Time Estimation
- 11:51Lessons Learned
- 13:03Scene Level Predictions
- 13:48Feature Correlation
- 15:35Information Channels
related talks
15mTowards automated mapping of pedestrian path networks at scale
19mHow we leverage AI to build great maps at massive scale
15mComputational Mobility: Learning, Prediction, and Algorithms
20mRoute Optimization Engine for the Real World
07mUrban Map Inference by Pervasive Vehicular Sensing Systems with Complementary Mobility
41mPredicting the Traffic Jam Congestion Aware Routing
36mClassification of Satellite Imagery
20mAI on the Fast Lane: Smart Traffic Systems to Unclog our Cities