Computer Vision Based Detection and Localization of Potholes in Asphalt Pavement Images
📜 Abstract
Asphalt pavement distresses have significant importance in roads and highways. This paper addresses the detection and localization of one of the key pavement distresses, the potholes using computer vision. Different kinds of pothole and non-pothole images from asphalt pavement are considered for experimentation. Considering the appearance-shape based nature of the potholes, Histograms of oriented gradients (HOG) features are computed for the input images. Features are trained and classified using Naïve Bayes classifier resulting in labeling of the input as pothole or non-pothole image. To locate the pothole in the detected pothole images, normalized graph cut segmentation scheme is employed. Proposed scheme is tested on a dataset having broad range of pavement images. Experimentation results showed 90 % accuracy for the detection of pothole images and high recall for the localization of pothole in the detected images.
✨ Summary
Research contribution
The paper presents a hierarchical computer-vision pipeline for asphalt-pavement images. Images are first converted to grayscale and represented using Histograms of Oriented Gradients (HOG), then classified as pothole or non-pothole images with a Naïve Bayes classifier. Images classified as containing potholes are processed with normalized graph-cut segmentation to estimate pothole locations and contours. Evaluation on a 120-image pavement dataset used 50 images for training and 70 for testing. The reported detection results were 90% accuracy, 86.5% precision, and 94.1% recall. Spatial localization was reported as successful for the pothole occurrences, while area-based comparison showed more limited agreement in some images. The average processing time for a pothole image was reported as 0.673 seconds.
Subsequent influence
The most clearly documented influence is a direct research continuation by several of the same authors. Their 2018 study, “Visual analysis of asphalt pavement for detection and localization of potholes”, cites and extends the 2016 approach by replacing HOG and Naïve Bayes with SIFT-based bag-of-visual-words features and an SVM classifier while retaining graph-cut-based localization. That later study reported 95.7% accuracy for pothole-image identification and 91.4% localization accuracy. (sciencedirect.com) Later review and application-oriented publications continue to cite the 2016 paper as an early, conventional computer-vision method for pothole detection, including a 2024 review of vision-based pothole detection and a 2022 smartphone pavement-management application. These citations indicate continued use as a methodological baseline, but do not establish that the exact HOG–Naïve Bayes–graph-cut implementation was adopted in an operational road-maintenance system. (pmc.ncbi.nlm.nih.gov)