PYOLO: A Plus YOLO Model for Occlusion Pedestrian Detection

Pedestrian and vehicle detection is an important task in computer vision and plays a vital role in applications such as autonomous driving, intelligent transportation systems and urban safety monitoring. Achieving high detection accuracy while maintaining computational efficiency is critical, especially when deployed on edge devices with limited resources. Although the YOLO series has achieved significant success in object detection, its performance tends to degrade when dealing with complex targets. To address this limitation, this paper proposes a novel model, PYOLO (PlusYOLO), designed to overcome the bottleneck in detecting more complex objects. We propose the SPPAN (Spatial Pyramid Path Aggregation Network), a pioneering cross-scale feature fusion method that ensures exceptional detection accuracy, even in challenging environments. Additionally, we propose the CSPPF(Spatial Pyramid Pooling Network) module, which improves the integration of multiscale target information. To enhance the model's efficiency without compromising detection performance, we incorporate the lightweight ContextGuided module, which significantly reduces the model's parameters and computational complexity. Finally, the SEAM attention mechanism is introduced to compensate for the lost responses to occluded targets, further improving the model's robustness in real-world scenarios.