Hybrid deep learning models for road surface condition monitoring

The goal of road surface condition (RSC) monitoring is to ensure transport safety and driving comfort. Motion sensors from a smartphone are usually used for RSC monitoring. These sensors transform the vehicle’s vibration into time series data. This work proposes two hybrid deep learning models for the classification of road surface anomalies: (1) Convolutional Neural Network (CNN) combined with Gated Recurrent Units (GRU) and (2) CNN-LSTM that combines CNN and Long Short-Term Memory (LSTM). In addition, we present a novel data labeling technique based on TCP/IP sockets that enables us to label sensor data in real-time using a smartphone application. Furthermore, a combination of Fourier and wavelet transforms is used as input to train the classifier models. In our experiments, the CNN-GRU achieved better performance compared to the CNN-LSTM model. [Display omitted] •Two hybrid deep learning models to classify road anomalies.•Two smartphones, for collecting sensor data and labeling in real-time.•Smartphones communicate via TCP/IP sockets.•Using a combination of FFT and DWT as an input to the classifier provides a superior classification.