UZH develops hybrid camera for greater road safety

Daniel Gehrig and Davide Scaramuzza from the Institute of Computer Science at the University of Zurich(UZH) have developed a system that combines a camera inspired by the human eye with artificial intelligence (AI). Current camera systems for driver assistance in cars take around 30 to 50 images per second. According to a press release, the new UZH system can detect obstacles a hundred times faster and with less computing power.

A standard camera takes 20 images per second, which are processed by an artificial neural network. This is trained to recognize cars and pedestrians. The data from the event camera is coupled with another type of AI system that is suitable for analyzing 3D data that changes over time. What the event camera sees is used to anticipate what the standard camera recognizes in order to increase its performance. "The result is a visual detector that can recognize objects just as quickly as a standard camera that takes 5,000 images per second," Daniel Gehrig is quoted as saying. All it needs is the bandwidth of a standard camera with 50 frames per second.

The method could significantly improve the safety of automotive systems and autonomous vehicles. "This way, they don't have a blind spot between the individual images - they can recognize obstacles more quickly," Davide Scaramuzza, head of the Robotics and Perception research group, is quoted as saying. The weakness of the system: things that move slowly could be overlooked. In addition, the images could not simply be converted into the usual data form for training the AI algorithm. If cameras with lidar sensors were used for optical distance and speed measurement in the future, the method could be further improved. "Such hybrid systems could be decisive in enabling the necessary safety for autonomous driving," says Scaramuzza. ce/js