Smart cars can be made better and safer through driver authentication powered by biometrics technology, according to a recent study that proposed a new technique that captures and identifies a driver's eye movements with a novel machine-learning algorithm.

In a paper published Feb. 6 in Image and Vision Computing, a team of researchers from the University of Toronto, in collaboration with the Alcohol Countermeasure Systems Corp., investigated the possibility of integrating low frame rate infrared cameras into smart cars for continuous in-car driver authentication, a relatively new area of research in biometrics. 

The study is the first to evaluate eye movements "in the wild" for biometric authentication, according to the authors. Other types of biometric identifiers, such as fingerprints, facial patterns, iris recognition and voice recognition, all use body measurements and calculations related to human characteristics to authenticate a person's identity. 

The researchers determined that eye movements and blinking behaviors display patterns unique to every person, and that using them as a biometric identifier can be very successful in authenticating people as they drive a car. The study proposed a model that combines blinking behavior with eye movement features "to build a robust multi-modal biometric system" that showed a high recognition accuracy. In the trials run for the research, the model achieved correct recognition rates up to 98.5% and error rates as low as 0%.

Technologically advanced cars already on the market integrate machine learning and computer vision techniques to improve the driving experience, including collision detection systems and autonomous driving. But a continuous driver authentication feature can make it safer to ride in special vehicles such as cash trucks, military vehicles and police cars or on public transport, the authors said. This technology would add a layer of security to cars that only authorized people are allowed to drive, but an accurate and reliable system has not yet been perfected.

"Our paper introduced a new technology that can be integrated into various biometric systems that rely on different biometric modalities," Sherif Seha, a Ph.D. candidate at the University of Toronto and lead author of the paper, told The Academic Times. "Eye movement and blinking behaviors inherit subject-unique features that are successful to a great extent in identifying drivers in a wild environment, even with low frame rate [infrared] cameras."

Multimodal biometrics are a combination of biometrics that increase security, accuracy and flexibility compared to using a single biometric, such as capturing face and eye traits at the same time for greater fraud protection. 

For the task, the researchers mounted infrared cameras, which create an image using infrared radiation, in the same way that an ordinary camera forms an image using visible light, on the dashboard of a driving simulator. A sample of 55 adult participants from Canada watched videos on a large screen of a driving session set in downtown Toronto from the driver's point of view. 

The sample was split into two groups, and each had their eye movements and characteristics captured by a different low frame rate eye-tracking system as the participants watched the five-minute driving session clips. A low frame rate camera captures fewer frames per second, which is less intensive for the camera. This type is typically more affordable and used in situations without fast-moving objects. 

The eye-tracking systems captured eye blinking behavior, including blinking pattern, speed and acceleration; eye movements, including gaze, fixations and saccades, or rapid eye movements between fixation points; and periocular features, which are physical traits surrounding the eyeball, including the eye-opening height, width and axial ratio.

Low frame rate devices can only extract behavioral eye movement characteristics, and are not typically capable of characterizing the physiological traits of the eye, such as the tissues and the muscles generating eye movements, according to the paper. 

Seha and the research team proposed extracting the blinking behavior and the periocular features along with the eye movements for a more powerful combination that they hypothesized would improve the overall recognition rates. They also developed a machine-learning algorithm that aligned eyes in each recorded frame of the infrared camera to estimate the area within the eye, which is one method of registering eye blinking behavior. The system evaluated and measured both driver identification and verification, which are types of authentication. 

"While conventional biometrics traits like facial characteristics, fingerprints and iris can be employed for driver authentication applications, most of these traits are intrusive and require the drivers' cooperation to provide their biometric trait," the authors said.

For example, fingerprints can be used to start a car, but for continuous authentication while the vehicle is in motion, the driver would need to touch a fingerprint sensor every few minutes. The authors emphasized that eye movements are the perfect way to verify a driver's identity because people exhibit unique behavioral and physiological patterns in their eye activity.

By tracking the combination of the eye movements, the eye blinking and the periocular features, the study achieved high recognition rates of up to 98.5% and error rates as low as 0% in the two modes of authentication. Seha said this makes their proposed system comparable to other biometric systems that use conventional biometric traits such as fingerprints. The rates are also considered reliable for various biometric applications, including for use as continuous driver authentication in smart cars.

However, the system did not perform as well as existing state-of-the-art eye movement biometric systems, achieving lower correct recognition rates in the identification mode and higher equal error rates in the verification mode. But the authors explained that they purposefully used low frame rate eye-tracking devices, which do not perform as well as cameras with higher frame rates, because they are more cost-effective. The researchers wanted to explore cheaper alternatives to the expensive eye biometric systems already on the market.

The current study also demonstrated a more realistic and free-movement experiment setting by using the driver simulation with real driving session videos. Previous studies have used expensive and high-accuracy eye-tracking devices, according to the paper, requiring participants to place their head on a chin-rest that restricted their overall movement.

Seha said the team's new proposed technology and its achieved results are promising, but the system is still in early stages as a proof of concept. It also has potential beyond integration in vehicles, including continuous identity authentication during online examinations, which have increased in use for virtual learning during the COVID-19 pandemic. The researchers plan to run further experiments that will integrate the technology in real cars being driven, scaling up from the video simulations.

The study, "Improving eye movement biometrics in low frame rate eye-tracking devices using periocular and eye blinking features" published Feb. 6 in the Image and Vision Computing journal, was authored by Sherif Nagib Abbas Seha and Dimitrios Hatzinakos, the University of Toronto; and Ali Shahidi Zandi and Felix J.E. Comeau, the Alcohol Countermeasure Systems Corp.