Investigation of the Road Surface Defects Using Robust Techniques with Multi-Sensor Integration

Road surface identification is crucial for enhancing the transportation system since developing new roads in poor countries is extremely challenging due to financial problems. Hence, it is more important to repair existing roads than to build brand-new ones. At the moment, manual inspections are used to find damaged roads, which is very expensive. A robust embedded device is required to identify rough road surfaces utilizing a multi-sensor. A few methodologies are examined in this study to determine surface roughness and categorize various road faults. The first approach uses autocorrelation and linear prediction coding (LPC) to calculate the road's roughness index (RI) using a filter-based strategy. The second Artificial Intelligence (AI) approach uses a common random forest (RF) approach to classify defects, including potholes and cracks. In the third strategy, the heatmap-based digital image processing (DIP) technique is used to classify road roughness into more precise categories, including potholes, rough patches, longitudinal, alligator, and transverse cracks. The YOLO model is used to validate defect detection results using a monocular camera, which shows excellent validation accuracy across all detection methods. The filter-based technique obtains an accuracy of 84.21% following YOLO validation. F1 scores of 78.4% for potholes and 83.9% for cracks are obtained using AI (RF) approach. After using YOLO for validation, the heatmap approach achieved 90.3%. The technology integrates geo-localization using GPS to determine the position of each problem, assisting in road maintenance.