RUP2S- YOLO: An Improved YOLOv8-Based Algorithm for Dense Pedestrian Detection
Juanmin Wang, Kun Huang, Jianyong Pi · 2024
Dense pedestrian detection is a key research direction in the field of computer vision, which plays a significant role in large-scale crowded public spaces. It provides strong technical support for security assurance, traffic management, and behavioral analysis in public areas. To address the issue of missed detection of small-sized occluded pedestrians in dense scenarios, this paper proposes an improved detection algorithm based on YOLOv8, named RUP2S-YOLO. For the backbone network, the RFCBAM module is used to replace the convolution module to enhance the capability of capturing subtle features and better extract the characteristics of small-scale pedestrians; for the C2f module, the UKMS module is designed as a substitution aiming to improve the model's capacity to process information of different scales; for the detection head, the P2S module is introduced for the accurate positioning of small-scale targets. Experiments on dense pedestrian datasets show that the improved RUP2S-YOLO outperforms the original algorithm in terms of performance, achieving an AP50 of 82.8%, and an AP50:95 of 55.8%, which represents an increase of 2.2% and 2.8% respectively compared to the original algorithm. Meanwhile, in comparison with other leading pedestrian detection models, the improved YOLOv8 algorithm presented in this article also demonstrates competitive advantages.