Abstract Manager

Traffic signal cabinets are the hub of intersection control and management. Equipped with sensor inputs, power, compute, control, and communication capabilities, in-cabinet technology has historically aimed to minimize delay, but a new application has enabled co-optimization for safety, especially for active modes.  

A new in-cabinet application was deployed on edge compute devices at 25 intersections in North America, to collect near-miss data for one year, resulting in over 200,000 hours of near-miss data. More than half of these intersections are in Canada. This effort was inspired by and seen as a natural extension of recent successes in short-term monitoring of near-misses that usually collect about 60 hours of data. The objective of this large scale pilot was to document lessons learned from a continuous safety monitoring approach.

The methodology employed first routed overhead camera feeds to a processor in the cabinet that used computer vision techniques to continuously extract road user trajectories. Pairs of trajectories were evaluated based on kinetic risk scoring techniques previously published for short-term studies, resulting in a time series of near-miss data covering all modes and possible collision configurations.

By shifting to an in-cabinet, continuous safety monitoring approach at these 25 intersections for an entire year, several key lessons emerged when compared to short-term monitoring. First, continuous monitoring offers a better ability to diagnose low frequency severe risk events such as those involving pedestrians, which can be missed in short-term studies.  Second, in-cabinet continuous monitoring offers a deeper ability to diagnose risk by correlating near-miss data to signal phase information. Third, continuous monitoring offers the ability to monitor trends over time. Fourth and finally, recognizing that computer vision is not perfect, a consensus on acceptable accuracy standards is needed before widespread deployment of this in-cabinet application.

The application, developed by Miovision, is not specific to any cabinet type or controller manufacturer. This presentation will give an overview of the application concept, the deployment to 25 cabinets, and the lessons learned from the first year of pilots.