Driver Activity Recognition Through Hand Gestures

Student thesis: Master thesis (including HD thesis)

  • Joakim Iversen
  • Tobias Hvass Mølbak
4. term, Software, Master (Master Programme)
Driver activity recognition and monitoring is a widely researched topic. Within the field, some of the most researched observations methods are eye and iris detection, which bases its prediction on the driver's gaze. However, fewer studies focus on driver's hand gestures to determine the activity. Current driver activity recognition systems are primarily focused on detecting certain states such as fatigue or sleepiness. These states are however only part of the problem that comes with the monotonous driving task and the accidents that follows.

Throughout this study, we research hand gestures and how these could be used to determine the current activity of a driver. The analysed hand gestures are the naturally occurring interactions that every driver is performing to control the vehicle, e.g. holding the steering wheel or using the gear stick. In order to do so, we are utilising a Leap Motion controller placed at the roof, just above the steering wheel. To prevent suboptimal driving styles, feedback is required to alert the drivers of their reckless behaviour. Most research papers experiment with feedback systems that provide feedback immediately, regardless of the situation. We develop a system utilising timely feedback, which provides feedback at appropriate times where the driver's cognitive resources can comprehend it, without compromising the driving performance.

The system averaged an accuracy of 85.60% in a performance test based on 17 evaluations performed in 2 cars with 13 participants. Our system showed good correlations between the actual situation and the system's predictions. For a field evaluation of the feedback part, a driving academy and 10 regular drivers evaluated the system. The system's feedback, not just timeliness but also frequency and message, was well received by both the participants and the driving academy, and some participants were surprised about their own subconscious behaviour. From this we conclude that hand gesture recognition is a viable method to accurately determine a driver's activities and that it should be studied further.
Publication date16 Jun 2017
Number of pages34
ID: 259745420