Detecting Abnormal Driving Behaviors by Smartphone Sensors Based on Multi-Feature Convolutional Neural Network

The abnormal driving behaviors detection is of great significance to reduce the incidence of traffic accidents and improve driving safety. However, there is not yet simple and convenient scheme to detect abnormal driving behaviors. To solve the problem, a novel abnormal driving behaviors detection method based on a new convolutional neural network and smart phone sensors is proposed in this paper. Firstly, the source data set is collected via the mobile phone gyroscope sensor and labeled according to the corresponding driving status. Secondly, multi-feature convolutional neural network(MFCNN) is proposed to extract and integrate features of data sets collected. Thirdly, appropriate network parameters are adopted to train MFCNN and obtain a stable model, which is an end-to-end network model. Finally, field test is conducted to verify the performance of the proposed scheme, which shows that MFCNN is superior to Random Forest and BP neural network. MFCNN achieved a 94% recognition rate for abnormal driving behaviors. The corresponding results for other two networks are 82% (Random Forest), 88% (BP neural network). Furthermore, MFCNN owns the smallest standard deviation of test results, which means that the network model of MFCNN is more stable and reliable.