Estimation of pothole structural parameters using depth data under oblique observation

Abstract Reliable quantification of road potholes is important for automated inspection and driving safety assessment. However, RGB-D observations collected from a vehicle under a forward oblique view often suffer from occlusion, scale variation, and depth noise. These issues make the results unstable and difficult to compare at different observation distances. This paper proposes a method for quantifying pothole geometry that utilizes a vertical profile along the wheel track. The profile describes the vertical variation of the pothole. A unified reference frame is built by normalizing the road reference plane. We also fuse profile information from different observation distances to reduce noise and local missing data, which improves the stability of the results. Experimental results show that the proposed method outputs stable key profile parameters. The RMSE of the fused profile error ranges from 2.3 % to 9.5 %, and overall performance is better than the profile extracted from a single-frame. This method provides a practical way to achieve interpretable quantification and comparative analysis of pothole geometry in vehicle-based RGB-D oblique observation conditions.