ドライバの姿勢変化に着目した安全確認行動の解析

抄録

We aim to construct a driver assistance system that is able to detect driver’s deviations. The system uses time-series head motion information during safety confirmation behaviors at non-regulated intersections. Therefore the system needs quantification of head postures. In this paper, we evaluate a head posture quantification method using one of the time-series data analysis method: Recurrent Self-Organizing Maps (RSOMs). RSOMs are added an ability of categorizing according to time changes to SOMs. Specifically, we discuss categorized results of safety confirmation behaviors at two intersections. A driving simulator is used for this experiment and we obtain driver’s upper body images. Input data is calculated with edge extraction and coarse graining using driver’s images. Number of output layer of RSOMs is 10. We compare graphs of categorized results of two intersections. From the results of this experiment, it is showed that same intersection graphs have similar trend, and this result means that there is same patterns in same intersection’s safety confirmation behaviors.