Intelligent road surface autonomous inspection

With the advancement of artificial intelligence, autonomous machines are featured with the ability to diagnose and assess the structural health of different systems. This paper presents a scalable mobile platform employed to autonomously and intelligently detect online small cracks on roads using a live camera feed and Artificial Intelligence (AI) methods. The robotic artifact is equipped with a vision-based localization system to enable autonomous navigation areas where GPS (Global Positioning System) may be poor or intermittent. The proposed approach runs at the edge a model of Convolutional Neuronal Networks (CNN) based on the Resnet 18 architecture to classify the image feed between cracks and those without cracks after training them with a combination of two public data sets and a data set generated in-house. The mobile robotic platform is scalable, depending on the particular context and requirements of the application. As opposed to off-line assessment tools, experimental results show the real-time capabilities of the system to autonomously navigate and detect cracks on a pavement structure with an accuracy of 95%.