Innovation Magazine July-August 2019

uninterrupted power supplies, which keep traffic signals connected and operating, even during lengthy and widespread power outages. Fifty Bluetooth sensors deployed last year throughout the city’s northwest provide refined information. The system detects Bluetooth-enabled devices on the roads, encrypts the data and routes it to the TMC, where a computer tracks the speed and travel time of the detected devices as they move through the city. By comparing real-time speeds to historical speeds, the system alerts staff when road links are travelling slower than normal and traffic signals may need to be adjusted and the public alerted to changing traffic conditions. Staff can act quickly when city emergency services are dispatched to a motor vehicle collision. During a recent demo for Innovation , staff sprang into action when an alert interrupted the proceedings. Graeme Cross, P.Eng., Surrey’s Traffic Signals Team Lead, pointed to a video screen showing a snippet of data from Emergency Dispatch. Moments later, a staff person panned a camera towards

the affected area, revealing a minor fender-bender and a pair of slightly dazed drivers—one of whom was still on the phone with 911. Another TMC staff person sent out a Twitter alert about the accident and adjusted traffic-light timings to ensure traffic snarls were minimized. The team accomplished all this several minutes before emergency crews had even arrived on scene. Cross said that Surrey was experimenting with machine learning, which can automatically gather and process vehicle and pedestrian data which can be used to apply real-time traffic management strategies as well as detailed road safety studies. He says that Surrey has a video archive of about 4,000 vehicle collisions, but currently the only way to gain helpful safety information from them is for a human to watch the videos and annotate them. “One of the things I’m excited about is the possibility of building a machine-learning neural network that can identify what a collision looks like,” said Cross. “If it knows what a collision looks like, it can identify what kind of vehicles were involved, the weather conditions, the

I N N O V A T I O N

J U L Y / A U G U S T 2 0 1 9

2 9