Automatic calibration of dead reckoning algorithms on mobile end devices (Masterarbeit)
Over the past years several companies have started to offer solutions for pedestrian navigation. One of the greatest challenges in this domain is to localize the user in GPS-denied environments, e.g. within buildings. Apart from absolute positioning methods e.g. using Wi-Fi signals, it is also possible to track a user's relative position by means of pedestrian dead-reckoning. This requires knowledge of both the heading and the velocity of the moving user in order to estimate the travelled distance and relative position, respectively. These quantities can be inferred from the inertial sensors of a mobile device; however, a central prerequisite is an accurate model for step length estimation. This thesis proposes and examines several methods to automatically calibrate the parameters of step length estimation models that depend on the person-specific body and walking characteristics. Two user studies were conducted to investigate the dependencies between these parameters and to gather input data for a calibration approach using machine learning techniques. The results show, that step length estimation based on walking characteristics is not satisfying. By using the body characteristics for the semi-automatic calibration the step length estimation is improving, which makes pedestrian navigation possible. The results of this thesis aid in the improvement of positioning methods based on pedestrian dead-reckoning and provide an enhanced understanding of step length estimation.