UK councils are using artificial intelligence to detect traffic accidents by monitoring CCTV camera footage. The aim is for the AI to alert traffic operators about incidents in real time, allowing them to act quickly.
This is one of the first uses of AI with public CCTV cameras to collect transport information in the UK, says Richard Cartwright of FlowX, a company working with the councils on the pilot schemes.
The system is trained to recognise nine different types of road user, including pedestrians, cyclists, cars and trucks. It will track their motion as they move through each camera鈥檚 field of view.
Advertisement
This month, Devon County Council has begun to use the algorithm to monitor live footage from 10 existing cameras. It will flag anomalies like slow or stationary vehicles, or vehicles travelling in the wrong direction, as potential incidents for an operator to investigate.
Currently, traffic operators rely on reports from police or bus drivers. These can be delayed by 10 to 15 minutes after an accident, says Cartwright.
Once launched, the algorithm will be given time to learn the regular traffic patterns at the location of each camera, so that it can recognise normal variations throughout a day, such as congestion during rush hour.
A pilot involving one camera has also been agreed with Leeds City Council and FlowX is in talks to use the technology across the UK, including with five other local authorities.
False flags
One challenge in creating the system has been minimising the number of false incidents flagged by the algorithm.
鈥淚f you have three very similar vehicles driving along the road at the same time, you could get a slow or wrong way vehicle detection, because the software thinks that one vehicle in one frame is a different vehicle in the next frame,鈥� says Peter Mildon at technology firm Vivacity Labs, which is involved with the project.
If there are too many false flags this could dissuade operators from using it, he says. However, reducing the system鈥檚 sensitivity may mean that some incidents are missed.
Similar AI-powered traffic monitoring systems have been used in China, such as the City Brain project in Hangzhou run by e-commerce giant Alibaba. Privacy concerns have been raised about projects like these, because the technology can track individual road users as they move through a city.
Identifying details of individual pedestrians or cars won鈥檛 be recorded in the UK pilots. The software only records the path a vehicle takes along the road, what kind of object it is 鈥� a car or cyclist, say 鈥� and a timestamp. No video footage is being saved by the algorithm, says Mildon.
Although the pilot algorithm won鈥檛 use number-plate or face recognition, law enforcement and insurance companies may be interested in doing so in the future to identify problematic drivers before an accident, says Emmeline Taylor at City, University of London.
鈥淲e know that algorithms are not perfect,鈥� she says. 鈥淭here are real and genuine concerns regarding the prejudicial profiling of drivers.鈥�
