|dc.description.abstract||This thesis deals with pedestrian localization by way of multi-sensor fusion. A special focus is made on the use of a foot-mounted inertial measurement unit (IMU) and its incorporation into other pedestrian navigation systems. Mounting to the foot location is preferred due to the anticipated dynamics of the foot during normal walking motion, but poses difficulty when fusing with other body-worn sensors. One of the main challenges lies in the non-rigid relation between navigation sources. This thesis approaches the problem by characterizing the human gait during walking motion and detecting instances in which spatial relations can be made. Navigational information of a pedestrian is often provided as a relative state measurement, such as step length, walking pace, rate of turn, etc. The fusion of multiple relative state measurements is non-trivial and requires manipulation of standard sensor fusion frameworks. Therefore, this thesis discusses certain frameworks for processing multiple relative state measurements in detail.
Two algorithms are presented for fusing a foot-mounted IMU with other body-worn relative state measurement systems. The first algorithm operates in a cascade architecture, which allows for easy implementation while still showing promising results. Drawbacks to the cascade architecture are discussed, and a second, centralized architecture is presented that addresses these issues. To serve as an example, a particular problem is addressed: fusing measurements from the foot-mounted IMU with a chest-mounted visual odometry system. By analyzing the human gait through the raw IMU signals, a relation is made between the position and orientation of the chest-mounted camera and the foot-mounted IMU. Using a high precision motion capture system, the human gait is analyzed and motion profiles of a typical step are calculated. Using these motion profiles as a basis, a simulation environment is developed to replicate visual odometry and foot-mounted IMU measurements and the navigation algorithms are applied to simulated data. Conclusions are drawn from simulation on the effectiveness of the respective algorithms and experimental data validates these findings. Experimental data is collected with an open source stereo visual odometry system and a MEMS grade IMU. In post-process, the experimental data is fed through the developed algorithms, and the results are compared to those found in simulation. The work presented in this thesis will inform the reader of the characteristics of a foot-mounted IMU solution and establish a methodology for fusing with any general pedestrian navigation device.||en_US