L.Y. Yao, W. Yang, W. Huang

Pages: 91-100

Abstract
Since driving behaviour has a close relationship with driving safety, forming good driving habits and avoiding unsafe driving activities are important for drivers to keep away from traffic accidents. However, abnormal driving behaviour detection system seldom appears on vehicles because of detection accuracy and many other reasons. In this paper, a novel method based on Kinect sensor is proposed to detect abnormal driving behaviour in real time. The kernel of the method is two-fold. A Kinect sensor is used to capture the joints information of the monitoring driver. Only 6 joints information are captured to form a driving behaviour descriptor. Then, an SVM is used to classify the behaviour into right driving behaviour or abnormal driving behaviour. We built a simplified car cockpit and test our method on a popular computer game, “Need for speed 11”, the experiment results show that the proposed method can effectively detect abnormal driving behaviour in real time.
Keywords: unsafe driving action detection; depth image; RGB-D data analysis