Attention-Enhanced YOLOv8 for Accurate Pedestrian Detection and Count Estimation

Abstract

Autonomous vehicles (AVs) are crucial for improving road safety by reducing human errors, which are a leading cause of accidents. They also offer significant potential for increased efficiency in transportation, lowering emissions, and enhancing mobility for the elderly and disabled. However, there are still significant obstacles that need more study to be overcome. For example, in crowded metropolitan settings, AVs must correctly sense their surroundings in order to operate safely. Numerous research papers explore effective methods for precisely determining the surroundings of AVs. However, there are still challenges to detecting non-static objects, especially pedestrians. This paper employs deep learning algorithms to achieve real-time pedestrian detection, thereby advancing AVS development. By enhancing feature extraction within the detection algorithm, In this article, we present the CBAM-YOLOv8 model, which integrates three CBAM modules into the backbone network of YOLOv8. Additionally, we introduce the DUA-YOLOv8 and ECA-YOLOv8 models. The Experimental results on a combined dataset of 1520 images collected from the INRIA and ETH datasets indicate that CBAM-YOLOv8 achieved higher precision, with slight improvements in recall, mAP@0.5, and mAP@0.5:0.95. The evaluation metrics for CBAM-YOLOv8 showed greater enhancements compared to SE-YOLOv8 and ECA-YOLOv8.