AI for Complete, Equitable Streets (ACES): Using Computer Vision and Machine Learning to Deliver Data-Driven Guidance for Complete Streets

Gaps in jurisdiction-wide, street-level infrastructure, traveler behavior/safety, and context data inhibit advancement of “Complete Streets” (CS) - safe streets that accommodate all users. Existing data collection methods are either not scalable (e.g., manual collection), expensive (e.g., deploying sensors jurisdiction-wide), non-representative (e.g., smartphone data), or incomplete or too aggregated (e.g., federal datasets). This Phase I proposal (ACES) uses 130 visual machine learning (VML) models to generate street-level data on 50 Infrastructure and 88 Context features, and vehicle, bicyclist, and pedestrian mix and counts (building on our 119 existing VML models for 127 built environment features). We apply VML to extract this, one-time-point, Infrastructure, Traveler Behavior, and Context data from anonymized street-level images, collected via an image capture prototype, mounted on Southeastern Pennsylvania Transit Authority (SEPTA) buses (in conjunction with SEPTA) and deployed on a minimum of 50% of the bus system. Using this and other relevant data, we produce a novel machine learning (ML) model to predict street-level travel mode mix and counts, jurisdiction-wide for three time points. Backup image sourcing mitigate project risk. ACES produces a comprehensive dataset to inform CS Siting, Prioritization, Design, Implementation, & Evaluation and a scalable, expandable CS data collection methodology.