Some federal and city transportation agencies are beginning to repurpose AI-equipped traffic cameras from tools associated with enforcement into systems designed to detect crashes faster and flag high-risk conditions in near real time. Recent pilots and field tests in Nevada and Florida, along with a related city hazard-detection rollout in San Jose, California, show measurable gains in incident detection and response speed by feeding live camera and sensor data through analytics. The shift reflects a practical bet that preventing secondary crashes and reducing exposure time after incidents can deliver more public-safety value than relying solely on tickets after the fact.
Nevada Pilots Cut Response Times and Find Hidden Crashes
The clearest evidence for this pivot comes from Southern Nevada, where a federal deployment evaluation found that an AI detection system uncovered 20 percent more crashes than had been captured through traditional reporting methods. The same program reduced average crash response times by nine to ten minutes, a gap that matters because secondary collisions often occur in the minutes after an initial wreck when lanes remain blocked and drivers are unaware of stopped traffic ahead. The system pulls from real-time data sources and applies predictive analytics to flag crashes and dangerous conditions as they develop, rather than waiting for a 911 call.
A separate pilot near Las Vegas pushed the concept further by layering AI on top of existing cameras, sensors, and connected vehicle data. That program contributed to a 17 percent reduction in primary crashes, partly by enabling targeted enforcement presence at high-risk locations and faster incident response when collisions did occur. Both Nevada results are reported in U.S. DOT ITS Knowledge Resources deployment evaluations, which helps distinguish them from vendor-funded case studies. Together, the two pilots suggest that the same camera infrastructure cities already own can be redirected toward prevention without large new hardware investments.
How Near-Miss Math Powers Prediction
The analytical engine behind these systems traces back to the Federal Highway Administration’s Surrogate Safety Assessment Model, which established a method for quantifying crash risk by studying near-misses rather than actual collisions. The model defines conflicts using two key metrics: time-to-collision, or TTC, and post-encroachment time, known as PET. According to the FHWA validation report, the default threshold for classifying a conflict is a TTC of 1.5 seconds, meaning any encounter where two vehicles are less than 1.5 seconds from contact gets flagged as a potential crash precursor.
Modern AI camera systems apply this logic at scale. Instead of a researcher manually reviewing hours of intersection video, computer-vision algorithms process feeds from dozens of cameras simultaneously, tagging near-misses in real time and building heat maps of where conflicts cluster. The original FHWA methodology was designed for offline simulation, but its core measures now run inside live traffic management platforms. That transition from academic model to operational tool is what allows agencies to act on risk patterns before a fatality forces a redesign. For drivers, the practical effect is that an intersection with repeated close calls can trigger signal retiming, speed advisories, or patrol deployment days or weeks before a serious crash would have occurred under the old reactive approach.
Drones and City Streets Expand the Toolkit
The camera-based approach is not the only technology gaining traction. A system called DARTS, tested on I-75 in Florida, combines drones, thermal imaging, and deep learning to detect and verify incidents from the air. In a field test, DARTS detected and verified a rear-end collision 12 minutes earlier than a local transportation management center staffed by human operators. That kind of speed advantage could matter on high-volume interstates, where delays in detecting and clearing incidents can increase the risk of follow-on crashes. The findings are described in a preprint posted to arXiv, a repository operated by Cornell University and supported by member institutions and donor funding. It represents an early but concrete demonstration that aerial AI monitoring can detect and verify certain incidents faster than a human-monitored center in the tested setting.
On city streets, San Jose has tested AI cameras for a different but related purpose. The city’s road safety pilot achieved 97 percent accuracy in detecting potholes, shifting maintenance operations from a model that depended on resident complaints to one driven by proactive detection. While potholes are not crashes, the underlying logic is the same: use continuous AI monitoring to find hazards before they cause harm. San Jose’s phased rollout shows how cities can start with lower-stakes applications like pavement damage and build toward more complex crash-prevention use cases as the technology matures and public trust develops.
The Enforcement Legacy Still Shapes Adoption
One reason this shift has been slow is that AI traffic cameras carry political baggage. For years, automated enforcement systems were associated primarily with revenue generation, and public backlash led several states to ban or restrict red-light and speed cameras entirely. A study published in May 2025, led by Dr. Aaron Cheng, Assistant Professor of Management at LSE’s Department of Management, examined how AI-powered enforcement affects road safety outcomes. The research points to a broader academic effort to distinguish between cameras that punish and cameras that protect, but the public perception gap remains wide.
That perception problem creates a real obstacle for agencies trying to deploy prevention-focused systems. If residents assume any roadside camera exists to generate tickets, political support for AI safety programs can evaporate before results arrive. The Nevada and Las Vegas pilots offer a counterargument: measurable crash reductions and faster response times that benefit all road users, not just the city budget. Advocates for prevention-first systems argue that pairing transparent performance metrics with clear legal limits on automated ticketing can help rebuild trust, allowing AI traffic cameras to be seen less as silent police officers and more as an always-on safety net for drivers, cyclists, and pedestrians.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
