Driving Behavior Tracking and Recognition Based on Multisensors Data Fusion

Driving behavior tracking and recognition are essential for traffic safety. This paper proposes a driving behavior monitoring and analysis method based on motion capture and artificial intelligence techniques. Different driving actions can be identified as normal or abnormal driving behaviors. The real-time motion is monitored by multiple miniature inertial measurement units (IMUs). Gradient descent method is used to fuse the raw data and update the driver's attitude information constantly. The body's joint angle series can be obtained by the iteration operation of the consecutive segment under the assumption of rigid structure. In order to accurately identify two arms' maneuvers under different traveling routes, we compare the traditional machine learning-based method with the proposed deep neural network-based approach using joint angle series. The recognition rates of both methods are above 99% in the experiment, and the algorithm performance in real scenario is also satisfactory. These results show that joint angle-based driving behavior recognition is an effective method, and it can be applied for driving training or guiding the novice.