Abstract
Pothole detection is one of the most important tasks for road maintenance. Computer vision approaches are generally based on either 2D road image analysis or 3D road surface modeling. However, these two categories are always used independently. Furthermore, the pothole detection accuracy is still far from satisfactory. Therefore, in this paper, we present a robust pothole detection algorithm that is both accurate and computationally efficient. A dense disparity map is first transformed to better distinguish between damaged and undamaged road areas. To achieve greater disparity transformation efficiency, golden section search and dynamic programming are utilized to estimate the transformation parameters. Otsu's thresholding method is then used to extract potential undamaged road areas from the transformed disparity map. The disparities in the extracted areas are modeled by a quadratic surface using least squares fitting. To improve disparity map modeling robustness, the surface normal is also integrated into the surface modeling process. Furthermore, random sample consensus is utilized to reduce the effects caused by outliers. By comparing the difference between the actual and modeled disparity maps, the potholes can be detected accurately. Finally, the point clouds of the detected potholes are extracted from the reconstructed 3D road surface. The experimental results show that the successful detection accuracy of the proposed system is around 98.7% and the overall pixel-level accuracy is approximately 99.6%.
Introduction
Potholes represent significant structural failures that compromise road safety and increase vehicle maintenance costs. Traditional road inspection methods are manual, making them slow, subjective, and potentially hazardous for personnel. While computer vision offers an automated alternative, existing methods often struggle. 2D image analysis is sensitive to lighting and shadows, while 3D modeling can be computationally intensive or sensitive to sensor noise. Moreover, most stereo-vision systems assume a perfectly level camera, whereas real-world vehicle "roll" (tilting) distorts data, making it difficult to distinguish flat road from damage. This paper introduces a robust algorithm that bridges 2D and 3D approaches for accurate and efficient pothole detection.
Methodology
The proposed system utilizes a multi-stage pipeline to process stereo image pairs into a 3D representation of the road surface.
- Disparity Transformation: The process begins with a dense disparity map. To compensate for vehicle tilt, the authors employ Golden Section Search (GSS) and Dynamic Programming (DP) to estimate the roll angle and rotate the disparity map, aligning the road surface for easier modeling.
- Undamaged Road Extraction: Otsu’s thresholding is applied to the stabilized map to segment potential undamaged road areas from obstacles or potholes.
- Road Surface Modeling: Disparities in undamaged areas are modeled as a quadratic surface using least squares fitting. To ensure robustness against outliers like gravel or noise, surface normal information is integrated, and RANSAC (Random Sample Consensus) is employed.
- Pothole Detection: Potholes are identified by comparing the actual disparity map to the ideal modeled surface. Significant deviations indicate damage, which is then extracted as 3D point clouds.
Key Insights: Modeling in Disparity Space
A critical innovation of this work is performing modeling directly on the disparity map rather than converting to a 3D point cloud immediately.
Immediate conversion to 3D involves an inverse depth calculation ($1/d$) that can amplify small sensor errors. By remaining in "disparity space," error remains linear and manageable. Furthermore, rotating the disparity map to account for roll creates a "flattening" effect. A road surface that appears as a complex curve in raw data becomes a simple, predictable parabola (v-disparity), significantly improving computational efficiency and accuracy. Additionally, using surface normals as a filter allows the algorithm to discard noise by identifying the distinct geometric signatures of actual road cavities.
Experimental Results
The system was evaluated using new pothole datasets (67 stereo image pairs) and benchmark datasets like EISATS and KITTI Stereo. Compared to state-of-the-art algorithms by Mikhailiuk & Dahnoun and Zhang et al., the proposed system achieved:
- Detection Accuracy: 98.7%, compared to 73.4% and 84.8% for baseline methods.
- Pixel-Level Precision: An overall accuracy of 99.6%, demonstrating high precision in defining pothole boundaries.
- Efficiency: The disparity transformation stage takes approximately 142 ms, moving the system closer to real-time deployment requirements.
Conclusion
This research presents a highly accurate and efficient method for pothole detection by combining disparity transformation with robust road surface modeling. While currently using some manual thresholds, future work may incorporate deep learning to automate parameter selection for diverse road types. Additionally, the authors plan to explore segmenting roads into localized planes to better handle complex urban geometries that deviate from a standard quadratic surface model.
Summary updated