San José's AI Road Safety Conditions Pilot — 97% Accurate Pothole Detection from City Vehicle Cameras

San Jose's Information Technology Department, working with the Transportation Department, launched the Road Safety Conditions Pilot in December 2023 to test whether AI models running on cameras mounted to city vehicles could automatically spot roadway hazards such as potholes, debris and obstructions in bike lanes and sidewalks.

The pilot aimed to detect road hazards automatically from footage already being collected by vehicles circulating on city routes, avoiding the cost of dedicated survey vehicles.

Detection hardware was piggybacked onto street sweepers and, in a later phase, parking-enforcement vehicles already doing their normal rounds. San Jose is a founding member of the GovAI Coalition, a network of local and state governments (including jurisdictions in California, Minnesota, Oregon, Texas, Washington and Colorado) that share AI tools and, eventually, training data, so a defect pattern learned in one city can in principle improve detection in another. A second phase, partly funded by a Toyota Mobility Foundation grant, is expanding camera coverage from a single district to a wider study area.

City staff published results showing the system identified potholes with 97% accuracy and trash or debris with 88% accuracy, based on footage collected from vehicles already circulating on city routes.

The published figures are city-reported pilot results rather than an independent third-party audit, and the program has not yet reported systematic before/after reductions in repair backlogs or costs - an important gap for future evaluation.