When you are driving, the app is linked to a camera that is “looking” carefully at your face, especially your eyes. If you get drowsy, when driving, the app literally wakes you up! It raises an alarm. There is more to its kindness. After waking you up, it sends you information about the nearest garage, hotel or guest house. It says, “You are just 5 km away from a resting place.”
If you didn’t know, this is an application of Artificial Intelligence (AI). Welcome to the world of the Fourth Industrial Revolution as National University of Lesotho (NUL) takes the lead.
The app is called “Driver Drowsiness (Sleepiness) Detector.”[If you want to experience the magic of this app, just visit the Pioneer Mall in Maseru and see the app among the on-going NULISTICE 2020 exhibitions. The expo ends on Friday the 24th of January, 2020].
This app is developed by two NUL students; Boitumelo Ntilane and Thabo Khabele. The gurus are supervised by Mr Napo Mosola and Mr Molete Sekese.
What motivated these guys? They have realised something quite troubling.
In the fast-paced hustles of modern life, many people are tired. But the same tired people often find that they have to drive. When you are on the wheel, alertness is key. “However, too many accidents have been caused by people who sleep behind the wheel,” Ntilane said.
Well, the police will arrest you for being drunk when you drive. But they can’t arrest you for being drowsy. Yet you are just as dangerous (if not more) when you are drowsy as when you are drunk. Their app seeks to address that dilemma.
So this is how the app works.
Smart cameras are placed in-front of you—the driver. Anyone who has seen smart modern cameras knows that they can be very small, so small that you won’t even notice them. The cameras face the driver. Although they see the whole face, they take a particular focus on the driver’s eyes. The app has been trained to identify the eyes on your face.
Yes, it won’t confuse your nose or your mouth for your eye— it knows exactly what your eyes are and how they look like. But humans have different faces and eyes. That is true. However, how do YOU know that a face is a human face even though humans have different faces? It is because behind those differences, there are startling similarities. We are all humans after all.
The app focuses on those similarities.
“We trained the app to identify all kinds of human eyes, even the Chinese-Japanese-Korean eyes, which have some unique features,” said Ntilane. “We used more than 2000 faces of people of all races to train our app.”
So when you start a car, the app opens. And the camera starts looking at you—the driver—and it does so every second of your driving period. It is looking specifically for how long you close your eyes.
But humans are always blinking, in which case they close eyes temporarily. So closing your eyes because you are blinking is the least of the app’s worries.
The problem comes when your eyes closes for an extended period. That extended period could be 4 seconds. Then the app realises that something is wrong. You are getting drowsy. You are falling into sleep. At that moment, an alarm rings, “wake up!”
How do they achieve this feat? It’s a bit complicated but we will endeavour to explain.
Your eyes can be characterized by something they call aspect ratio. Aspect ratio is often described as a proportional relationship between the width and height of an object. While an eye’s width remains constant, its height varies with blinking.
When our eyes open, the aspect ratio is around 0.31. When they close, the aspect ratio approaches zero. So the app is constantly looking at the eye and computing its aspect ratio. If the ratio is around 0.31, all is fine. If the ratio is approaching zero, the question becomes, for how long?
If it approaches zero for a moment, then the app dismisses that as blinking. However, if it approaches zero for some time, say 4 seconds, then, again, someone is getting drowsy and the alarm is triggered.
That is the first way to identify a drowsing driver. The second way is to watch the speed of blinking. If the eye were to blink too fast, say 3 times per second, that is a sign of dozing off. Then the alarm is triggered to alert you.
Now your questions.
What if it’s dark in the car? “We are going to experiment with cameras that can “see” in darkness,” answered Ntilane. Yes, such cameras exist. Another question—can the camera see through eye spectacles? Yes and No. Yes for transparent eye glasses; No for dark eye glasses.
Welcome to the world of the fourth industrial revolution.