High-frequency LED-headlight for faster pseudo image generation and object detection with event-based cameras

Event-based cameras, also known as Dynamic Vision Sensors (DVSs), are biologically-inspired sensors based on a fundamentally different operating principle than conventional cameras. Instead of continuous image acquisition, they asynchronously capture changes in brightness at the pixel level and output them as vents characterized by their position, timestamp, and the sign of the change. Due to their high temporal resolution and dynamic range, these sensors are increasingly important for object detection in machine vision, particularly in autonomous driving. Nevertheless, efficiently using fast and asynchronous event streams for precise and robust object detection with low latency remains a key challenge. Most previous approaches are based on the integration of asynchronously recorded events, to generate so-called pseudo-frames, which are then used for object detection. However, these methods are associated with relatively high latency times, as the accumulation of the events typically covers periods of at least 50 milliseconds. This paper presents an approach to minimize accumulation time using a high-frequency Pulse-Width Modulated (PWM) Light Emitting Diode (LED) headlight system. Modern automotive headlight systems feature high pulse-width modulation frequencies, allowing the projection of light distributions at high frame rates. Through these artificially generated brightness changes invisible to human drivers, the number of registered events within a time interval increases significantly. The contribution at hand describes a prototype combining a PWM LED headlight system with an event-based camera. In a first step data was recorded in static settings for various objects. The analyses show a reduction in the integration time required for the generation of pseudo-frames resulting in an improved perception of the surroundings. This approach opens up new possibilities for faster and more efficient object detection in the context of autonomous driving at nighttime.