VPCW-YOLO: An Improved YOLOv8 Algorithm for Vulnerable Pedestrian Detection under Complex Weather Conditions
Jian Su1,2,*, Jiaqi Wang2
1 School of Information Science and Engineering (School of Cyber Science and Technology), Zhejiang Sci-Tech University, Hangzhou, China
2 School of Computer Science, Nanjing University of Information Science and Technology, Nanjing, China
* Corresponding Author: Jian Su. Email: 
Computers, Materials & Continua https://doi.org/10.32604/cmc.2026.082531
Received 17 March 2026; Accepted 27 April 2026; Published online 29 May 2026
Abstract
Despite significant advances in object detection technology, vulnerable pedestrian detection in intelligent transportation systems remains highly challenging under complex weather conditions. Environmental factors such as fog, rain, and snow often lead to occlusion, motion blur, and low-contrast images, making small-scale or weak-featured vulnerable pedestrians difficult to accurately identify. Therefore, improving the detection accuracy and robustness of vulnerable pedestrians in complex weather scenarios has become an urgent research problem. To address this issue, this paper proposes an improved YOLOv8-based vulnerable pedestrian complex weather detection algorithm, termed VPCW-YOLO. The proposed method enhances detection performance through multiple structural optimizations. First, a C2f_ST module is designed by integrating Spatial and Channel Reconstruction Convolution (SCConv) with a triplet attention mechanism to strengthen feature representation and improve the model’s focus on critical regions. Second, a residual self-attention based (RSAB) module based on a self-attention mechanism is introduced to enhance global feature modeling capability under complex weather conditions. In addition, a Space-to-Depth operation is embedded into the backbone network to preserve more fine-grained information. Finally, a P2 small-object detection layer is added to improve the detection performance for distant and tiny pedestrians. Experimental results on the augmented BGVP dataset demonstrate that VPCW-YOLO achieves 73.0% mean Average Precision (mAP)@0.5 and 49.2% mAP@0.5:0.95, representing improvements of 5.4% and 5.7%, respectively, compared with the original YOLOv8 model. Generalization experiments on the Real-world Traffic Sign Detection in the Wild dataset (RTTS) pedestrian subset show that VPCW-YOLO achieves a 6.3% improvement in mAP@0.5. The results indicate that the proposed method effectively improves the detection accuracy and robustness of vulnerable pedestrians in complex weather scenarios while maintaining certain generalization capability, providing a promising solution for pedestrian safety perception in intelligent transportation systems.
Keywords
Intelligent transportation systems; object detection; pedestrian detection; attention mechanism
Open Access
Login
Register
Submit a Paper
Propose a Special lssue
Download PDF
Downloads
Citation Tools
