Driving Maneuvers Detection using Semi-Supervised Long Short-Term Memory and Smartphone Sensors

Abstract

Driving maneuvers detection is an important component of proactive traffic safety management and connected vehicle systems. Most of the existing studies used supervised learning concepts to train their models with labeled data. These methods achieved promising results but were limited by the heavy dependence on the labeled data. With the development of mobile sensing technologies, massive traffic-related data can be efficiently collected by mobile devices (e.g., smartphones, tablets, etc.). Considering the high costs of labeling data, this paper proposed a semi-supervised deep learning method to learn from the unlabeled data. Data from a smartphone’s accelerometer and gyroscope were collected by different drivers with a variety of smartphones, vehicles, and locations. Three long short-term memory (LSTM) models were trained with the proposed semi-supervised learning algorithm. Experimental results indicated that the proposed semi-supervised LSTM could learn from the unlabeled data and achieve outstanding results with only a small portion of the labeled data. Using much fewer labeled data, semi-supervised LSTM could achieve similar results compared with the supervised method. Moreover, the proposed method outperformed other machine learning methods (e.g., convolutional neural network, XGBoost, random forest) on precision, recall, F1-score, and area under curve. More and more traffic data will be available in the future, the proposed method is expected to make use of the undiscovered potential from the massive unlabeled data.