Drowsy drivers to get AI-assisted safety prompts

Fatigued drivers are one of the most common hazards on the road, but sleepy-heads on commercial wheels are to get AI-assisted safety prompts courtesy of biometric-tracking technology.

The tech companies are targeting vehicle fleet operators and in particular long-haul trucking firms, reports IEEE Spectrum, since such drivers usually drive the farthest distances and lengthiest times.

While automakers have employed camera-based systems to monitor biometrics, such as drivers’ eye movements, posture, breathing and hand placement, for inattentiveness, companies are now using machine learning to detect signs of drowsiness.

Driver monitoring tech developed by Samsara, Motive, and Nauto (all California-based) deliver real-time audio alerts to a drowsy driver, prompting them to take a break to help avoid fatigue-related accidents. Motive’s AI monitors yawning and head movements; Nauto’s tech tracks yawning, blink duration and changes in the driver’s body posture; while Samsara’s system tracks drowsiness symptoms such as excessive eye closure, head nodding, eye-rubbing, slouching, and yawning.

In order to develop such a system, Samsara had to train its AI on billions of minutes of video footage to come up with a model aligned with the clinical definition of drowsiness (the Karolinska Sleep Score). All the drowsiness-detection-tech companies have configured their systems so that fleet managers are directly contacted if a driver continues to operate the vehicle after they’ve been alerted of their drowsy condition.

While Samsara says it is not seeking mass adoption of its technology in consumer vehicles, auto manufacturers such as Ford, Honda, Toyota and Daimler-Benz have incorporated similar alert signals for drowsy drivers to take a break. But as vehicles with Advanced Driver Assistance Systems (ADAS) become more common, drowsiness-detection systems might turn into a feature of semi-autonomous vehicles as drivers could engage in risky behavior.

Diverse multimodal in-car biometric data sets now available

In other biometrics-related auto news, professor and senior researcher in biometrics security and privacy Sébastian Marcel announced a new “iCarB” data set.

The in-car biometric data set, which are actually three subsets, contains face videos, fingerprint images and voice samples for driver recognition collected by the Biometrics Security and Privacy Group at Idiap Research Institute. The data set features demographic diversity as there is a fifty-fifty gender split, skin colors across the entire Fitzpatrick-scale spectrum, and ages ranging from 18 to 60-plus among the 200 data subjects.

According to Marcel, the iCarB datasets can be used to evaluate and benchmark face, fingerprint and voice recognition systems; create multimodal pseudo-identities and to train and test multimodal fusion algorithms; create Presentation Attacks from the biometric data and to evaluate Presentation Attack Detection algorithms; investigate demographic and environmental biases in biometric systems, using the provided metadata.

The paper “in-Car Biometrics (iCarB) Datasets for Driver Recognition: Face, Fingerprint, and Voice” can be found here. The links to the subsets for face, fingerprint, and voice, respectively, can be found via Sébastian Marcel’s LinkedIn announcement here.

Article Topics

automotive biometrics  |  biometrics  |  dataset  |  Idiap  |  monitoring  |  real-time biometrics