Road Marking Damage Degree Detection Based on Adaptive Feature Fusion

The clarity of road markings is crucial for safe vehicle operation. Over time, road markings may detach to different extents, making it crucial to promptly assess the degree of damage and repair severely affected areas. To evaluate the degree of damage to road markings, we propose a method that integrates pixel segmentation and region segmentation. Pixel segmentation primarily segments damaged markings at the pixel level, while region segmentation aims to capture the entire region of the road marking image. However, indiscriminate feature fusion during segmentation may decrease the model's sensitivity to target boundaries and its capability to segment small objects. Consequently, we propose an attention-guided adaptive feature fusion method to enhance region segmentation. Experiments show that our model achieves a pixel accuracy of 0.9920 and an average intersection-to-merge ratio of 0.8601 in region segmentation, accurately assessing the degree of damaged markers. The method proposed in this paper is simple, effective, and holds great potential for real-world applications.