Real Time Recognition of Non-driving Related Tasks in the Context of Highly Automated Driving

With the continuous development and improvement of advanced driver assistance systems up to highly automated driving functions, the driving task is changing. There is no need for the driver to permanently supervise automatic driving functions of SAE J3016 level 3 and 4. The driver is allowed to engage in non-driving related tasks temporarily. However, if the automated vehicle reaches its limitations, the driver needs to react appropriately to a takeover request. Driver state monitoring systems might enable adaptive take-over concepts to support the driver in such situations. In order to recognize the currently performed non-driving related task by a technical system it is necessary to fuse different features from several measurement signals to infer the currently executed task of the driver. Main features of non-driving related tasks include the driver's visual orientation and position of his or her hands. In this paper, a methodology is presented to detect a non-driving related task using Hidden Markov Models to represent the temporal relationships of characteristic features. Measurement data was obtained from participants in a driving simulator and used to train and evaluate the presented system with various nondriving related tasks.