Multi-modal system to detect on-the-road driver distraction

The number of traffic accidents is increasing year on year. Studies show that distraction during driving is one of the major causes of traffic accidents. Lack of attention to the primary task of driving due to involvement in activities such as phone conversation, eating, texting, conversation with the co-passenger lead to serious injuries and fatalities. In order to reduce the traffic accident due to driver distraction, in this work, a monitoring system is developed using physiological, behavioral and vehicle signal. Motion signal (accelerometer and gyroscope), electrocardiogram (ECG), galvanic skin response and CAN-Bus signal were collected during the on-road driving session undertaken by 8 participants. Features were extracted from these signals. Feature space from each signal was evaluated independently to identify driver distraction. To improve the recognition accuracy the multimodal feature space was fused and evaluated. Since the high dimension of fused feature space suffers from the curse of dimensionality, feature selection techniques were applied to obtain optimal multi-modal feature space. An average accuracy of 99.85% was obtained when multimodal feature space was generalized using ensemble bagged classifier. An average accuracy of 99.1% was obtained when the multimodal feature space was reduced to 10-D space using the Relieff feature selection technique and generalized using ensemble bagged classifier.