Announcing SpaceNet 5: Road Networks and Optimized Routing

Determining optimal routing paths in near real-time is at the heart of many humanitarian, civil, military, and commercial challenges. This statement is as true today as it was two years ago when the SpaceNet partners announced SpaceNet Challenge 3 focused on road network detection. In a disaster response scenario, for example, pre-existing foundational maps are often rendered useless due to debris, flooding, or other obstructions. Satellite or aerial imagery often provides the first large-scale data in such scenarios, rendering such imagery attractive in the quest to aid disaster response.

Consequently, the SpaceNet partners are pleased to announce SpaceNet 5: Road Networks and Optimized Routing. This public competition will challenge competitors to automatically extract road networks from satellite imagery, along with travel time estimates along all roadways, thereby permitting true optimal routing.

Figure 1. Optimal paths in Khartoum. Left: Simple shortest distance path between start (green) and end (red) locations. Line widths are proportional to speed limit. Right: Path optimized by travel time, rather than distance. The very different paths taken in this plot illustrate the need to extract speed limit and route time estimates for all roadways.

Motivation

The automated extraction of roads applies to a multitude of long-term efforts such as: improving access to health services, urban planning, and improving social and economic welfare. This is particularly true in developing countries that have limited resources for manually intensive labeling and are under-represented in current mapping. Updated maps are also crucial for such time sensitive efforts as: determining communities in greatest need of aid, effective positioning of logistics hubs, evacuation planning, and rapid response to acute crises.

Using artificial intelligence (AI) to automatically extract road networks from aerial or satellite imagery at regional and city scales remains a surprisingly difficult yet critically important challenge. Entities such as Google Maps and OpenStreetMap (OSM) provide useful maps of the world, yet accurate road labeling on these platforms is manually intensive and often slow to update in dynamic scenarios. For example, following Hurricane Maria, it took the Humanitarian OpenStreetMap Team (HOT) over two months to produce a fully validated map of Puerto Rico, even with a team of thousands of volunteer mappers. Furthermore, large regions of the world remain poorly mapped.

Competition Structure

Current approaches to road detection from remote sensing imagery tend to focus on just the road pixels (e.g. 1, 2, 3), or topology (e.g. 1, 2, 3, 4). While these approaches are are a very useful first step, they they fall short of addressing automated routing, as they cannot be used directly to assess optimal routing since travel time estimates are lacking.

We will therefore ask competitors to output a detailed graph structure with edges corresponding to roadways and nodes corresponding to intersections and end points, with estimates for route travel times on all detected edges. Including estimates for travel time permits true optimal routing, not just the shortest geographic distance. This challenge will help fill the capability gap between semantic segmentation of roads and optimized routing for disaster response, among other scenarios.

Figure 2. Left: Road network for sample SpaceNet chip. Right: Satellite imagery overlaid with roads colored by speed limit.

The SpaceNet 5 challenge will run for two months, with an anticipated launch date of September 3, 2019. Data labeling for new SpaceNet cities is ongoing now, and will be publicly released as the competition launch nears. Evaluation will occur with a modified version of the APLS metric, tuned to optimize travel times between nodes of interest.

Stay tuned for more updates on SpaceNet 5, and in the meantime read up on those graph theory techniques.