Texting at the wheel detected by autonomous machine-learning system
Image credit: Dreamstime
Engineers at the University of Waterloo have developed machine-learning algorithms which can detect with high accuracy when drivers are texting or otherwise distracted while on the road.
While self-driving cars are being tested on public roads around the world, it will be years before these vehicles become widely commercially available. In the meantime, Canadian researchers hope that the development of new autonomous systems could contribute to improvements in road safety.
Researchers at the University of Waterloo have created an autonomous system capable of detecting actions which could suggest driver distractedness. This uses machine-learning algorithms – which can detect patterns of behaviour when trained on enormous datasets – to recognise the hand movements indicative of texting, talking on a phone, reaching into the backseat to retrieve an object and other actions which deviate from safe driving.
The system employs a suite of cameras to monitor the driver, in addition to the machine-learning software. It classifies the risky activities in terms of possible threat to road safety, based on the duration of the action and other factors.
According to Professor Fakhri Karray, director of the Centre for Pattern Analysis and Machine Intelligence at the University of Waterloo, this information could come in useful by warning or alerting drivers when they are dangerously distracted, improving road safety.
The addition of autonomous features to cars in the future could, he suggests, allow for protective measures to be triggered when the software detects signs of dangerous distraction.
“The car could actually take over driving if there was imminent danger, even for a short while, in order to avoid crashes,” said Professor Karray.
The Centre for Pattern Analysis and Machine Intelligence has dedicated previous research efforts to the automatic detection of other signs of danger on the road. This has included frequent blinking (a sign that the driver is in danger of falling asleep), head and face positioning and subtle physiological signals, such as pupil size and heart-rate variability, which could suggest that a driver is not paying adequate attention to the road.
The team of engineers hopes to combine the detection, processing and classification of these various signals of driver distractedness into a single automated safety system.
“[Driver distraction] has a huge impact on society,” said Professor Karray, citing reports which estimate that this is to blame for up to 75 per cent of all traffic accidents. According to the World Health Organisation, road accidents caused 1.25 million deaths in the year 2010 alone.