Lightweight YOLOv8 Pedestrian Detection Model Based on FasterNet

To address the challenges posed by the computational complexity of the YOLOv8 detection head, which limits its applicability on common surveillance devices and intelligent processing terminals, as well as the low accuracy and poor robustness of existing detection models in occlusion tasks, a lightweight pedestrian detection model based on an improved YOLOv8 was proposed. The backbone network was integrated with the FasterNet network, which requires less computational power and memory. Furthermore, a novel object detection head based on group normalization was introduced to reduce computational costs by reusing feature information in object detection tasks. Experimental results demonstrate that the proposed model achieves superior lightweight performance compared to the original YOLOv8n model. Specifically, the number of parameters was reduced by $\mathbf{4 5 \%}$, the computational cost was decreased by 39%, and the model size was reduced by 41%, with only a 0.1% decrease in detection accuracy. These results highlight the practicality and efficiency of the proposed approach.