DARKSIGHT: INTELLIGENT EMBEDDED SYSTEMS FOR NIGHTTIME PEDESTRIAN AWARENESS

Abstract

Nighttime driving poses significant risks due to reduced visibility, often leading to pedestrianrelated accidents. Traditional vision systems struggle to reliably detect pedestrians under lowlight conditions, necessitating intelligent and adaptive solutions. This study introduces DarkSight, an embedded night-vision system designed for realtime pedestrian detection using advanced image processing and machine learning techniques. The system integrates infrared imaging, convolutional neural networks (CNNs), and edge computing to identify and track pedestrians efficiently in dark environments. By embedding the detection algorithms into compact hardware, DarkSight ensures low latency, energy efficiency, and scalability for automotive applications. Experimental results demonstrate high detection accuracy across various low-light scenarios, outperforming conventional night-vision systems. The findings underscore the potential of intelligent embedded solutions in enhancing pedestrian safety, reducing accidents, and supporting the development of smart vehicle technologies.