A Centralized Approach to Pedestrian Localization Using Multiple Odometry Sources

A method is presented for pedestrian localization by use of stereo visual odometry and a foot-mounted inertial navigation system (INS). As opposed to prior approaches for fusing such systems, a centralized model combines the state vector for both visual odometry and INS mechanization using an Extended Kalman Filter (EKF) framework. Correlations are maintained in the centralized filter by pseudo-measurements relating the position and heading of the two systems. Gait monitoring is performed in order to detect vertical leg conditions in which the two systems can be related by a baseline separation. In addition, zero velocity updates are applied to foot velocity states to estimate INS errors. External measurements are applied through a standard EKF update, and their effect on filter performance is shown. Results are presented using both simulated and real data.