Multi-scale feature fusion with adaptive edge enhancement for robust sidewalk detection in urban environments

This paper presents a novel approach for sidewalk detection in urban environments using multi-scale feature fusion combined with adaptive edge enhancement techniques. The proposed method integrates a modified U-Net architecture with attention mechanisms and incorporates geometric constraints based on urban infrastructure characteristics. Our approach processes RGB images captured from vehicle-mounted cameras and pedestrian viewpoints to segment sidewalk regions with high accuracy. The multi-scale feature fusion module captures both fine-grained texture details and global contextual information, while the adaptive edge enhancement component refines boundary detection between sidewalks and adjacent surfaces. Experimental validation on a custom dataset of 5,000 urban images from various cities demonstrates that our method achieves a mean Intersection over Union (IoU) of 87.3% and an F1-score of 91.2%, outperforming existing state-of-the-art methods by 5.8% and 4.6%, respectively. The approach shows robust performance across different lighting conditions, weather scenarios, and urban layouts, making it suitable for real-world applications in autonomous navigation systems and accessibility planning tools.