Real-time pavement classification using an embedded artificial neural network

Abstract

The capacity to analyze pavement condition can be a very important feature for either automobiles or general road quality surveys. This information can be used as reference to decide when a road requires maintenance or as a means to select a different route when travelling, based on a driver requirement for comfort. There are methods available for general road analysis such as the Pavement Condition Index (PCI) that requires extensive manual survey of the pavement. In this context, this work presents a system that uses machine learning techniques for automatic pavement classification, based on accelerometer readings that are used to classify roads between two categories: asphalt or paving stone. A triaxial accelerometer and Global Positioning System (GPS) modules are used as peripherals to a real-time operating system (RTOS) that can execute high precision data acquisitions and also classify the road in real time. This operation requires that an accelerometer module is mounted next to the vehicles center of gravity and is calibrated to the vehicle mounting point. The data acquisition mode is used to obtain data in the city of Campinas, SP - Brazil, containing acceleration from longitudinal, lateral and vertical axis and geolocation data from the GPS. Then, this data is analyzed and an extensive feature selection process is executed to filter the best metrics that can be used for training the Artificial Neural Network (ANN) for pavement classification between the two types of road. The model uses features extracted from acceleration data in both time and frequency domains and achieved an accuracy of 94% in the test set. The model was added to the embedded system, allowing classification of the pavement in real time.