Application of artificial intelligence and object detection for the determination of defects in flexible pavement

Abstract

According to the Association of Victims of Traffic Accidents (AVIACTRAN), on average there are 10 potholes per kilometer of flexible pavement in the city of Lima. This is due to the lack of timely road maintenance by government authorities, as they do not have a system that allows them to identify different pavement defects in real time to make decisions. To address this issue, this article proposes the use of the Cascade Trainer GUI algorithm and Python programming to determine defects in flexible pavements. The proposal involves training the algorithm with images of different pavement defects using cell phone cameras or drones for data collection and pavement condition assessment. The implementation of the model provides a 60% saving in the time of detection of functional pavement defects compared to the traditional method. The methodology detects 5 types of defects (alligator cracking, edge cracking, block cracking, potholes, and weathering) with an accuracy of 70%. This innovative approach offers an efficient and fast solution for the management of road infrastructures in urban and rural environments.