Enhanced Road Crack Detection Using a Hybrid Feature Extraction Approach and Bayesian Optimized Support Vector Machine

Abstract

The integrity and maintenance of road infrastructure are critical for ensuring safe and efficient transportation. Timely and accurate detection of road cracks is essential to prevent accidents, reduce maintenance costs, and ensure the longevity of road networks. Traditional manual methods for road surface inspections are labour-intensive, prone to human error, and inconsistent. Recent advancements in machine learning and computer vision techniques have facilitated automated approaches for road crack detection, significantly improving accuracy and efficiency. This paper proposes a novel methodology for road crack detection by combining Histogram of Oriented Gradients (HOG), Grey Level Co-occurrence Matrix (GLCM), and Local Binary Pattern (LBP) for feature extraction. A combined feature set is then subjected to Neighborhood Component Analysis (NCA) for feature selection, ensuring only the most relevant features are retained for classification. The final classification is performed using a Bayesian Optimized Support Vector Machine (SVM) classifier, enhancing the model's predictive accuracy and robustness. The proposed method was tested on a comprehensive road crack dataset and achieved a remarkable accuracy of 98.94%, surpassing traditional models.