Driver Monitoring in Future VehiclesIntroduction Future vehicles will not only implement systems which assist drivers to safe driving, but they also include services which are used in the home and the office today. Internet connections, email services, navigation applications, and driver assistance systems are already in the market for luxury cars and they are expected to city cars in the year 2010. Such a trend should improve traffic safety and allow more freedom for the drivers. However, the information flow increases considerably which may too much attract the attention of the driver if the information flow is not scheduled. The vehicle manufacturers have pointed out the problem recently. They are jointly developing new innovations to schedule the delivery of information in pursuance of driver ability. AIDE [1] is an EU funded project which is developing a user interface that adapts to the ability of the driver to receive messages. 28 industrial and academic partners all round Europe are participating in the 4 years project, with a total budget of 12,5 MEuros. Driver Monitoring Driver Characteristics (DC), Driver Availability Estimator (DAE), Driver State Degradation (DSD), Traffic and Environment Assessment (TERA) and Cockpit Activity Assessment (CAA) are the basic modules to be designed for profiling the driver. The modules process information available from a camera, CAN bus, and a navigation system. VTT is working with the CAA module which is described here. The module is developed in close co-operation with Volvo Technology, Siemens VDO and Fiat. The aim of CAA is to adapt warnings or aids from the driver assistance systems to the driver's level of attention on the driving task. The DAE module is focused on the primary tasks (i.e. how much pedals, steering wheels, etc. are used). On the other hand, CAA estimates the driver's mental load caused by the secondary tasks, e.g. operating radio, phone calls, daydreaming, etc. DCAA module
The state estimation in CAA is based on the faceLab camera vision system from a company called Seeing Machines [2]. The framework is a commercial system, which is intended for tracking and acquisition of human facial features. Using 3-dimensional measurements from a stereo vision system (see. Figure 1), the distraction level is estimated with multidimensional classification algorithms.
Distraction can be divided to two main components: visual and cognitive. For visual distraction analysis, the cockpit is divided into individual clusters e.g. mirrors, radio, dashboard, etc. The result depends on the glance time and frequency of looking at certain cluster. Most of the time, the eyes of the driver are directed towards the road which is the main target, as shown in the histogram in Figure 2. The other sectors correspond to gaze and head directions caused by some external distraction which require more attention from the driver. Cognitive distraction is typically due to phone discussion, daydreaming or lack of vigilance. It causes long glance fixations to a certain point with small variation. Additionally, some standardised gaze based measurements will be applied for distraction estimation. As a result of sensor fusion with data available from the CAN bus of the vehicle, the performance of the driver is estimated. The result from CAA and other modules are fed to the user interface, which is adapted according to the driver availability and ability measurements.
 SENSATION and Further Development
VTT also participates in the SENSATION project [3], in which seat foil sensors are developed for driver monitoring, among other sensors, see Figure 3. Based on muscle movements, muscle relaxation and other physiological signals, fatigue, wakefulness and stress of the driver is estimated.
 The future vehicle sensing systems cannot rely on any single method; instead vigilance assessment will be the result of sensor fusion from several sources. VIRGIL, which is a newly proposed subproject for EU IP PReVENT, is not only intended for further developing the topics in SENSATION, AIDE and the earlier EU funded projects (AWAKE, SAVE, etc.), but we also attempt to improve the sensing capabilities and robustness of current driver monitoring methods, and develop new sensing principles. The new approaches utilise e.g. computer vision and measurement of body muscle relaxation, and focus more on post-processing of available data. Novel ideas are needed for merging the results of earlier projects to simplified practical driver monitoring modules which allow development of better in-vehicle warning systems.
|
