Driving Operation Recognition Using Smart Cushion Based on Deep Neural Network

With the increasing rate of traffic accidents, driving safety has attracted attentions of researchers from academy and industry. Nowadays, most of the traffic accidents are related to abnormal driving operations. This chapter proposes a novel method aiming to recognize the basic driving operations. The method uses smart cushion for data collection and deep neural network (DNN) algorithm for classification. In particular, the proposed method can recognize five driving operations: normal driving, stepping on the brakes, stepping on the accelerator, stepping on the clutch, rotating the steering wheel. A smart cushion that contains four pressure sensors is used to collect the driving-operations-related data. The defined processing workflow involves: (1) a preprocessing phase where a digital filter is used for data noise reduction; (2) a feature extraction phase based on a sliding time window technique able to extract time-based features; (3) a merging phase where features are combined together; (4) a phase where principal component analysis is executed before classification to reduce high-dimensionality of combined features; (5) application of the defined DNN for classification. Using our proposed DNN, basic driving operations can be therefore recognized. The experimental results show that the proposed driving operations recognition method achieves an average recognition accuracy rate of about 97%.