Embedded Night Vision System for Pedestrian Detection Using Deep Learning

Pedestrian detection in low-light and nighttime conditions is a critical challenge for road safety. Traditional vision-based detection systems often fail in poor lighting, increasing the risk of accidents. This project presents an Embedded Night Vision System for Pedestrian Detection Using Deep Learning, utilizing infrared thermal imaging and deep learning models to enhance detection accuracy in dark environments. The system integrates an ODROID XU4 microcomputer running Ubuntu MATE with an infrared camera for real-time pedestrian recognition. Advanced models like YOLO and Faster R-CNN are trained on thermal image datasets to accurately distinguish pedestrians from other objects, ensuring high efficiency with minimal false detections. Experimental results demonstrate high accuracy (93.5%), precision, and real-time processing capabilities (1.2s per frame), proving the effectiveness of deep learning over conventional methods. The system offers a cost-effective and portable solution for applications in automotive safety, surveillance, and industrial monitoring. Future improvements could involve multi-sensor fusion, real-time alert systems, and edge AI optimizations to enhance performance. This research contributes to making pedestrian detection more accessible and reliable, reducing nighttime accident risks and improving overall road safety.