Position Classification and In-Vehicle Activity Detection Using Seat-Pressure-Sensor in Automated Driving

Abstract:
Increasing the safety of vehicles, at all levels of automation, requires an accurate assessment of driver state and consequently improved comfort and awareness. Over the past decade driver state monitoring has been widely researched. EEG and ECG sensors, wristbands, and eye trackers are some of the commonly used devices for catching mental state of the driver and/or passenger in research and driving simulators. However, most of the implemented sensors are interfering and drivers are asked to wear the sensor while driving. This kind of data collection is not desirable in the automotive industry since it limits the feasibility of the proposed systems in future vehicles. In this contribution, a seat-pressure-sensor, BodiTrak2 Pro mat, on a reference chair is investigated. The reference chair is a model of driver seat with configurable seat and backrest inclination, which has rigid surfaces to reduce the complexity of seat deformation. To observe the driver state with the designed system a pilot study is conducted involving four subject with dispersed characteristics and the pressure distributions during various activities in different positions are collected. Then, machine learning algorithms are trained to classify the posture and activity of individuals. To demonstrate the trade-off between accuracy and cost, three sensor resolutions are utilized for the classification and the outcomes of all three are presented. The results show that the subjects' posture and activity can be classified using only the seat-pressure-sensor. The characteristics of the subjects affect the performance of the position classifiers, however determining effect of characteristics on the activity detection requires further investigation. The suggested monitoring method is applicable also in real vehicles as a different approach to estimate driver state.