Inertial/magnetic sensors based pedestrian dead reckoning by means of multi-sensor fusion

The challenges of self-contained sensor based pedestrian dead reckoning (PDR) are mainly sensor installation errors and path integral errors caused by sensor variance, and both may dramatically decrease the accuracy of PDR. To address these challenges, this paper presents a multi-sensor fusion based method in which subjects perform specified walking trials at self-administered speeds in both indoor and outdoor scenarios. After an initial calibration with the reduced installation error, quaternion notation is used to represent three-dimensional orientation and an extend Kalman filter (EKF) is deployed to fuse different types of data. A clustering algorithm is proposed to accurately distinguish stance phases, during which integral error can be minimized using Zero Velocity Updates (ZVU) method. The performance of proposed PDR method is evaluated and validated by an optical motion tracking system on healthy subjects. The position estimation accuracy, stride length and foot angle estimation error are studied. Experimental results demonstrate that the proposed self-contained inertial/magnetic sensor based method is capable of providing consistent beacon-free PDR in different scenarios, achieving less than 1% distance error and end-to-end position error.