Slashing Street Crashes With Artificial Intelligence

Last year my wife was hit and injured at an intersection. When the police showed up they said “no one is bleeding so we don’t need to write a police report.” What?! Overnight she went from being a marathon runner to having irreversible herniated & bulging disks in her neck and back which made it impossible to stand, walk, or sit without pain. Imagine if that was you. Last year crashes killed 631 and injured 63,740 across Los Angeles County. 248 pedestrians and cyclists lost their lives. Ultimately transportation agencies have to wait years for people to get injured and killed BEFORE there’s enough data to identify the root safety concerns, let alone solve them. Nationally 30% of crashes go unreported. Ped & Bike crashes are rare enough that even with 5 years of data there typically isn’t a pattern and when a crash happens you’re 20 times more likely to die than a vehicle occupant. No wonder more people don’t walk and bike in LA!

At Street Simplified, we specialize in helping public sector transportation agencies solve crashes using video analytics and AI. We’ve already worked at 70 intersections with 18 agencies including several in Los Angeles County: Pasadena, Arcadia, and Santa Clarita. We propose to use our analytics to study the 20 deadliest intersections across the county for pedestrians and cyclists. We use 24 hours of video taken from high-resolution temporary cameras to capture near misses, red-light running, speeding, and ped/bike non-compliance. In a single day, we capture 20 times as much risk data as you’d get from a year of crash reports. We then build detailed safety reports which outline the risk patterns we observe and provide a menu of proven mitigations that reduce crashes and save lives. The last step is to walk through the data and our safety analysis with the agencies that manage those intersections so they can take action on problems which in most cases, they weren't aware of or didn’t know how to fix. We already have a shortlist of locations extracted from TIMS, but the exact locations will depend on which agencies choose to participate.

Applying a proven model or solution to a new issue or sector (e.g., using a job recruiting software or strategy to match clients to supportive housing sites, applying demonstrated strategies from advocating for college affordability to advocating for housing affordability and homelessness, etc.)

We assess pedestrian-cyclist and vehicle safety risk and thus will use total crashes for the impact analysis. The top 10 crash locations in LA all had over 15 reported injury crashes last year. At 20 locations like this, that would be ~3000 crashes over 10 years. Typically 1 in 200 crashes results in a fatality so that’s 15 people dead over 10 years at these locations. If the ratio of total fatalities to ped/bike fatalities from last year persists, 5 or 6 of those killed would be pedestrians or cyclists. We recommend potential countermeasures backed up by years of research from FHWA which in aggregate typically would reduce crashes by 80%. We work closely with each agency partner to ensure that they take action and realize these benefits. In the worst case let’s just say they implement only half of our proposed mitigations. That’s still 1200 injury crashes eliminated, 6 lives saved, and 2 or 3 pedestrians or cyclists who can live another day over the next 10 years.

How we are measuring impact 1. We highlight potential recommendations which have already been proven. Crash reduction is measured with “Crash Modification Factors” which indicate the expected reduction in crashes for a given mitigation. The CMF clearinghouse http://www.cmfclearinghouse.org/ provides a list of over 1000 proven countermeasures that we draw from. Impact can be assessed by multiplying the current crash rate by the modification factor to estimate the statistical impact of a mitigation. 2. We conduct before/after evaluations using our video analytics to measure near misses, red-light running, speeding, and ped/bike non-compliance before and after making a change and this allows us to estimate crash reduction within weeks of a project rather than years. 3. Five years from now, the crash data will show a statistical decrease in crashes at locations where we studied and agencies took effective action.