IoT Based Road Surface Condition Monitoring Using Spatiotemporal Feature Learning

Internet of Things devices enable the monitoring of road surface condition (RSC) through a set of motion sensors that capture vehicle vibration data. These sensors’ data can be converted to a much more usable form, by utilizing data preprocessing techniques. In this context, deep learning (DL) models are suitable for road anomaly detection. This study aims to evaluate the effectiveness of a spatiotemporal feature learning approach in detecting road anomalies using a new RSC-IoT dataset. Therefore, based on a combination of DL models such as Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), and Convolution Neural Network (CNN), we propose three hybrid deep learning models: ConvLSTM, TimeDistributed-CNN-LSTM, and TimeDistributed-CNN-GRU. Moreover, in order to achieve better performance, the proposed road anomaly detection models are combined into two ensemble learning methods namely averaging ensemble and weighted average ensemble. The experimental results show that the weighted average ensemble method outperformed the averaging ensemble and individual models.