CSAI Phase I - Skylite Group

The foundation of any system that aims to effectively facilitate data-driven decisions relies on the quality of the data inputs into the system. For Phase I of Complete Streets Artificial Intelligence Initiative we propose an approach that focuses on the quality of the data, data attribution and data fusion. Significant advances in recent years in computer vision allow for accurate extraction of data from both overhead and surface based imagery. These datasets can be combined with GeoSpatial data in order to form a more complete picture of the situation on the ground. Anonymized and aggregated GPS tracks from Strava is a freely available dataset for infrastructure development purposes. This dataset would allow for rapid identification of unsafe road crossings, analysis of utilization of existing bike lanes, as well as areas where new ones might need to be established. For data fusion, we propose deploying a GeoLLM to build a system that would take a prompt like “show all crosswalks over 45 mph roads, with Strava pedestrian traffic during the last month and serious traffic incidents within 500m during last year” and convert it into queries across multiple data sources to display data with attributes on the map.