Research on Tourist Flow Detection in Scenic Areas Based on Improved YOLOv8

With the continuous development of tourism, the accurate monitoring and management of tourist flow in scenic areas have become increasingly important. This study aims to achieve precise detection of tourist flow in scenic areas by improving the YOLOv8 model. We propose an improved model based on the YOLOv8 framework, named LSC-YOLOv8. This model uses the Swin Transformer as the backbone network, and adopts the XIoU loss function and BiFPN for multi-scale feature fusion, thereby significantly enhancing detection accuracy. By comparing the performance of five different models: YOLOv8n, S-YOLO, B-YOLO, X-YOLO, and LSC-YOLO, the results show that LSC-YOLO performs best on all evaluation metrics, with an accuracy of 0.895, a recall of 0.842, mAP50 of 0.923, and mAP50-95 of 0.683, indicating significant advantages in detection accuracy and robustness. These results suggest that LSC-YOLO is the optimal choice and provide important references for further optimization of other models. The proposed model significantly enhances the detection capabilities for both people and vehicles, making it particularly suitable for daily flow monitoring in scenic areas, thereby improving management efficiency and ensuring tourist safety. This research provides reliable data support for the management and planning of tourist attractions, offering significant practical application value and promotion potential.