Accuracy of Hybrid Models of Detection, Classification, and Quantification for Automatic Road Damage Evaluation

Automatic identification of road damage conditions using available technology is extortionate. High-accuracy detection classification of road damage type and amount can provide more accurate information about detailed damage based on digital imagery. Relevant stakeholders can allocate resources more efficiently and improve the accuracy of costs associated with handling road repairs by increasing the accuracy of the type and volume of road damage detection. This study aims to explore road damage detection models by identifying and qualifying precisely the variety of damage and calculating the volume of damages. This study used a quantitative exploratory approach. The scope of this study is asphalt road pavement, which has seven types of damage. The initial stage in the methodology is the process flow of preparation, labeling, and training datasets, followed by an analysis of performance measurement data, the accuracy of detection, and the classification of pavement damage. Thenit continued with the analysis of performance measurement data and the accuracy of calculating the volume of each significant damage. The resulting hybrid model contributes to ORACE: Originality, Reliability, Accuracy, Completeness, and Efficiency in identifying road damage. The object detection model has achieved excellent precision performance, a high precision value (95.1%), and can detect objects more precisely. As many as 80% of all positive objects are identified, and the model has a good balance between recognizing objects with high precision and capturing most objects that should be detected (high sensitivity). Meanwhile, the quantification of the volume of asphalt road damage (e.g., potholes), where the level of accuracy is determined based on the comparison of the volume of calculated data to reference data, is 97.89%. Accuracy shows that the pothole volume calculation process application model can start an excellent calculation of road damage volume.