Japan develops new technology to help self-driving cars recognize traffic lights 150 meters away

As autonomous vehicles continue to develop, they must be equipped with systems that work in harmony with existing infrastructure and need to be able to handle complex urban landscapes. For example, they need to handle traffic lights of different shapes, sizes, and mounting locations; multiple city sign colors; and arrow lights that indicate directions. In order for autonomous vehicles to operate safely, they must be able to identify such objects at a distance, understand them, and react accordingly in a very short time.

(Image source: Kanazawa University)

To this end, according to foreign media reports, researchers at Kanazawa University in Japan have developed a new method that can reliably help autonomous vehicles deal with complex traffic conditions in cities and different arrows indicating directions. The system allows vehicles to recognize traffic lights 150 meters away and slow down or adjust driving operations smoothly and naturally.

Autonomous vehicles face special challenges in urban areas where there are dense crowds of vehicles, pedestrians and various objects. They usually need to use sensors to estimate their position and use high-precision digital maps that can provide the precise location of stationary objects to handle this situation. These technologies can help vehicles estimate their own location and the surrounding situation, so as to change their position or speed more intelligently. In all situations, successfully dealing with traffic lights is crucial.

"In order for self-driving cars to react safely, they must be able to see traffic lights at least 100 meters away," said Ryo Yanase, co-author of the study. "So we created a unique algorithm that uses digitized maps to identify the relevant traffic lights and then calculates the necessary acceleration or deceleration maneuvers. We also found a way that makes self-driving cars better than most people at discerning distant arrow lights."

(Image source: Kanazawa University)

Using a high-resolution lens or telephoto lens to identify a traffic light from 100 meters away seems like a relatively simple matter, but when quick decisions need to be made and a wide field of view becomes critical, processing time may increase and the field of view may be reduced. The algorithmic solution based on the digital map created by the researchers works with sensors to map the position of the vehicle and the "region of interest" containing the traffic light as the vehicle approaches the intersection.

The system does not require excessive resources, so the vehicle's central processing unit (CPU) can process it in real time. The researchers applied the method to various urban driving conditions to see how it works. They found that the system can easily identify traffic lights even 150 meters away and adjust the speed of the car accordingly.

"We succeeded in creating and verifying two valuable results," said Keisuke Yoneda, the study's first author. "First, we were able to detect and respond to traffic lights at long distances with almost no false positive detections, while still managing CPU load. Second, our system can recognize small arrow-like objects, even those with less than 10 pixels."

The above research results will help develop the necessary technology to enable autonomous vehicles to more safely navigate busy urban areas filled with other vehicles, pedestrians and many objects.