Driver Head Movements While Using a Smartphone in a Naturalistic Context

International audienceSmartphone usage while driving is a dangerous activity directly linked to 17% of deadly accidents in France in 2015. While it can potentially impact every road user, professional drivers are perhaps the most affected collective as they spend a high amount of time in their vehicles on a daily basis, whether they deliver goods or transport people. Detecting smartphone usage is interesting for many reasons, from legal controls to forensics investigation in accidents, without forgetting the possibility of alert the driver of the danger he is engaging in. In this study we evaluate the pertinence of using driver head rotation movements to automatically predict smartphone usage at the wheel. In order to fit the system to the particularities of professional drivers, a naturalistic driving study have been conducted. 15 operational vehicles from two private French transport companies were equipped with near-infrared cameras, embedded real-time computer vision and machine learning techniques. Common behavioural patterns have been revealed. Head rebounds are a constant (letting the driver switch gaze between the road and the device). Additionally, the duration a driver spends looking down from a reference neutral direction can be used as a reliable parameter to predict smartphone usage. On the other hand some divergences between van and bus drivers head movements have been noticed. A real-time smartphone usage detection system has been implemented from the results of this study. Preliminary results are encouraging and a prototype of the system is already being tested by professional drivers in a naturalistic context