Sub-Optimal Head Position Classification Using OpenVINO and SVM

Improving our understanding of passenger-driver behavior during roadway driving is crucial for advancing the development of effective autonomous vehicle technologies. Understanding passenger behavior will allow for the development of appropriate control structures that would be able to tune responses to prevailing passenger preferences. This area, however, proves to be a challenge despite the prevalence and continued emergence of high quality technical products, because, we argue, there continues to be an incomplete understanding of how to enable machine interpretations of real-time passenger responses for such tuning behaviors. Using machine learning techniques and optimization algorithms to create a support vector machine (SVM) classifier, this paper investigates a method for using remotely-sensed passenger head position during drives in an attempt to understand the passengers’ behavior in relation with, and perhaps as a window into their judgements of, the driver and/or their behavior. Our results demonstrate that remote head position sensing is feasible and may be useful even at lower fidelity (trinary) classification levels, especially if resolvable in near realtime. In the case of this exploratory work, we observed that most of the passengers maintained a straight head position throughout the drive, which we take as validation of the potential for future applications using evolved versions of this or similar technologies, for instance, in leveraging such inferences across larger groups of vehicle occupants to enhance prediction accuracy.