Human state estimation during crashes

Nowadays, there is a trend towards vehicle safety systems that are adaptive, e.g. the adaptive load limiter or smart airbag. Conventional restraint systems have typically one level of operation, and this level is a compromise over several loading conditions and occupant types. The effectiveness of occupant protection systems can be significantly improved when the loading condition and the actual status of the occupant are known and when this information is used to adaptively control the restraints. Most of the interesting states, like the actual forward position of the chest and chest compression, are, however, complex or expensive to measure; especially during a crash. Hence, this paper describes a method to estimate these states using a model based observer: a Kalman filter. Firstly, a number of sensors is selected that provide a useful input to the Kalman filter, and that can be easily installed in a vehicle. Next, a manageable but accurate occupant model is proposed, as a system model is essential to the Kalman filter. This model is derived from a Madymo hybrid III dummy model, and consists of 11 bodies, including the seat belt. Validation results of this model against the original Madymo model are shown. Then, the method to setup the Kalman filter is discussed and applied to the occupant estimation problem. Estimation results are obtained with the sensor data from the original Madymo model, and from real-world crash tests. These results show that relative chest position and compression can be well approximated with the proposed estimator system. Moreover, an application is presented, in which the human state estimator is applied to a controlled seat belt system, and results are shown. Special attention is paid to the fact that the estimator should be able to run in real-time, but some adaptations may be needed to achieve real-time execution. Finally, an outlook is given on different applications of the human state estimator.