Pedestrian Detection Using Event Cameras and YOLOv8: An Optimized Event Stream to Event Frame Conversion Algorithm

Pedestrian detection is a crucial aspect of intelligent transportation systems and autonomous driving technologies, ensuring the safety and reliability of these systems. This paper presents a novel approach to pedestrian detection utilizing event cameras and the YOLOv8 model. The core of our methodology lies in a newly optimized algorithm for converting event streams to event frames, specifically tailored for pedestrian recognition. By addressing the unique challenges posed by high-speed and dynamic environments, our approach enhances the accuracy and efficiency of pedestrian detection systems. The proposed algorithm leverages the temporal and spatial resolution advantages of event cameras, effectively reducing noise and improving the clarity of the event frames through advanced denoising techniques such as temporal filtering, spatial filtering, and polarity consistency checks. These techniques ensure precise feature extraction and robust pedestrian identification. We conducted extensive road tests using the EVK4 event camera, capturing dynamic scenes involving pedestrian movement in various real-world conditions. Our experimental results demonstrate significant improvements in detection performance, achieving an Average Precision (AP) of 88.6%, a detection speed of 114 FPS, and high robustness under low light and high contrast conditions.