Pedestrian-vehicle detection based on an improved YOLOv8

With the rapid development of autonomous driving technology, precise detection of vehicle and pedestrian targets is crucial for ensuring driving safety. This paper improves upon YOLOv8, the latest generation of real-time object detection frameworks, by introducing a dual attention mechanism composed of Global Average Pooling (GAP) and Squeeze-and-Excitation (SE) blocks to enhance the model's ability to extract pedestrian features and improve detection accuracy. The Global Average Pooling operation allows the GAM to capture global contextual information of the image, while the SE block adaptively recalibrates the feature map using channel attention mechanisms, highlighting information more critical to vehicle and pedestrian detection. Combining these two attention mechanisms not only improves the model's recognition capabilities for pedestrians of different sizes and shapes but also enhances detection stability in complex environments. Extensive testing on a self-collected and annotated traffic dataset demonstrates that this method significantly outperforms other advanced methods in terms of accuracy, recall rate, and average precision. The findings of this study provide strong technical support for the safe implementation of autonomous driving and point to new directions for future research.