Improvement of inertial sensor based indoor navigation by video content analysis

Foot-mounted inertial systems for indoor positioning and pedestrian guidance are an elegant and cheap solution to track first responders within buildings and underground structures. They can run completely autonomous, i.e. they do not require preinstalled infrastructure installations. But there are some difficult problems to solve: Starting from a known position the positioning error increases with the travelled distance, if only a double integration of the accelerations is performed. Therefore it is necessary to cut down the integration intervals and to reposition the system from time to time vis-à-vis a known landmark. Several algorithms have been developed at TU Graz to reduce these errors. The precise recognition of the motion patterns allows for performing zero velocity updates (ZUPT) to achieve a high accuracy in distance. Further improvements are gained by the fusion of additional sensors like barometer or GPS. Different map matching algorithms are used to perform periodic repositioning. Since ground floors often are not available, HSLU has developed a video content analysis based repositioning method that increases the accuracy of the heading of the IMU system considerably.