Trunk roll constraint to improve heading estimation in pedestrian dead reckoning navigation systems

A method for mitigating the heading error growth in personal dead reckoning systems using inertial measurement units is described. From data collected during laboratory tests on human subjects to characterize the biomechanics of normal walking, four kinematic parameters that identify (or are signatures of) the beginning and ending of turning motion are identified - trunk roll, trunk yaw, foot yaw and stride width. Observers for estimating these parameters are described. These parameters are used as flags in the dead reckoning algorithms to identify actual heading changes from apparent heading changes caused by inertial sensor output errors. Therefore, during straight line walking (when heading changes are equal to zero) the algorithm estimates inertial sensor errors. When heading changes are identified, the sensor bias estimation stops and the inertial sensors are integrated in an open loop fashion to generate an estimate of heading. Experimental results show that this approach can result in a heading error reduction between 6% and 60% compared to simple open loop integration of inertial sensor outputs.