12.06.2024Bio-Inspired Cameras and AI Help Drivers Detect Pedestrians and Obstacles Faster

Recognizing obstacles – especially pedestrians – quickly and reliably is crucial for any driver-assistance system. Every millisecond counts when a car is moving at high speeds and a child runs across the street. However, traditional RGB cameras are comparatively slow as they either suffer from low framerates (missing events that happen between two frames) or, in the case of high framerates, generate too much data to evaluate quickly.

Using a clever combination of a regular RGB camera and a so-called “event camera” (which excels at recognizing pixel-precise changes, e.g., speed) Daniel Gehrig and Davide Scaramuzza from the IfI’s Robotics and Perception Group were able to construct a visual detector that recognizes potentially dangerous situations a hundred times as fast as the best driver-assistance camera systems available today. Gehrig (who also won the UZH Annual Award for his PhD Thesis about this work) and Scaramuzza published this groundbreaking method in Nature. Congratulations!

 

Daniel Gehrig, Davide Scaramuzza. Low Latency Automotive Vision with Event Cameras. Nature. 29 May 2024. DOI: 10.1038/s41586-024-07409-w