Prediction of Pavement Roughness using a Deep Neural Network Approach

One of the most important variables in pavement performance management is the International Roughness Index (IRI). Conventional IRI measurement techniques depend on costly instrumented cars and skilled personnel. The IRI updates on pavements are limited by equipment and labor expenses associated with traditional measurement models. This papers implements a unique imaging-based Deep Neural Network (DNN) model that directly determine the IRI values from pavement photos. This model demonstrated that, commonly utilized vertical accelerations or 3-dimensional (3D) images, it is possible to identify IRI using 2-dimensional (2D) images. The fast development of photographic equipment has made it little and convenient, lightweight sports action cameras, like GoPro Hero series, to combine with electronic image stabilization devices and record smooth, high-framerate videos. When compared to existing methods include Artificial Neural Network (ANN), and Radon Transform (RT), this implemented DNN method is achieved high values of 0.95 of precision in Midwest Transportation Center (MTC) dataset, and 99.6 of R ^2 of long- term pavement performance (L TTP) dataset.