A Survey of Deep Learning Approaches for Pedestrian Detection in Autonomous Systems

Pedestrian detection is a critical component in autonomous systems, ensuring safe navigation and collision prevention. With the rise of deep learning, object detection frameworks like the YOLO (You Only Look Once) series have achieved notable success in identifying pedestrians across varying environments. This survey presents a comprehensive review of YOLO-based models, emphasizing the capabilities and limitations of the current YOLOv8 architecture. While YOLOv8 offers improved detection speed and accuracy compared to its predecessors, it still faces challenges in detecting partially occluded, small-scale, and densely grouped pedestrians. To address these issues, we propose three enhanced models—YOLOv9, YOLOv10, YOLOv11 and YOLOv12—each integrating architectural advancements such as adaptive attention mechanisms, multi-scale feature fusion, and temporal context modeling. Through extensive analysis and benchmarking on standard pedestrian datasets, we demonstrate that the proposed models significantly outperform YOLOv8 in terms of detection accuracy, precision, and robustness. This work aims to serve as a foundation for future research in pedestrian detection for autonomous systems.

Keywords: Pedestrian Detection, Autonomous Systems, Deep Learning, Object Detection, YOLOv8, YOLOv9, YOLOv10, YOLOv11, Feature Fusion, Attention Mechanism, Temporal Modeling