Smartphone assisted pedestrian localization within buildings

Location based services are becoming an indispensable part of our life. The wide adoption of satellite based positioning - Global Positioning System (GPS) has practically solved the problem of outdoor localization for a wide range of scenarios. Unfortunately, satellite based positioning is not possible indoors because of weak radio signals and loss of the direct line of sight from the satellites. Therefore, significant efforts have been motivated towards finding a practical solution for indoor localization especially in regards to localizing pedestrian. Certainly the topic of indoor pedestrian positioning does not lack research, there have been several research studies also various commercial solutions have been developed. What is common for all of them is that no approach has yet made a big impact within this area (e.g. GPS for outdoor localization). The reason behind this is that they either need an expensive infrastructure deployment (e.g. Wi-Fi access points) or have specialised hardware needs (e.g. network card), or have low accuracy and low reliability or have privacy issues such that pedestrians’ location is continuously monitored without their consent. There is also a trade-off between accuracy and cost. Sensing infrastructures (e.g. Wi-Fi) involving higher investments provide better accuracy where as those involving lower investments (e.g. QR codes) provide lower accuracy. Even worse, systems could not logically localize a pedestrian that is whether they are on this room or the adjoining room separated by a dividing wall and somehow if they do, they require large amounts of infrastructure to be installed into the environment. Smartphones are little less to ubiquitous. Thus, this thesis investigates an alternative approach to indoor pedestrian localization that uses smartphones to provide accurate, reliable, low cost logical localization. A significant emphasis is given on user privacy and minimal usage of infrastructure or none at all. It is demonstrated that how the information from smartphone sensors can be used for positioning in an infrastructure free environment by means of a case study. An extension to the well-studied inertial navigation technique is implemented using smartphone mounted on a toy vehicle over an artificial testbed – Scalextric track. Having learnt that infrastructure free positioning is possible using only the inertial navigation sensors embedded in smartphone, off the shelf stride estimation methods (foot step detection techniques and stride length estimation models) are applied to investigate the most suitable stride estimation method for smartphone based pedestrian dead reckoning (PDR) positioning system. Unfortunately, what was most noticeable in all the methods was that their performance was user specific and importantly, dependent on heuristic parameters. In addition, the position error grows overtime because of slowly accumulating errors in the measurement of inertial sensor. To reduce the dependency on heuristic parameters we investigate the statistical approach – ‘Kalman filter’ to get a better estimate of the stride lengths. Nevertheless, drifts are mitigated by enforcing constraints from map using map matching technique – multiple uncertain routes engine (MURE). MURE is an extension to the Kalman filter that allows location to be described using multiple discrete Gaussian distributions bound to a map. The developed map aided pedestrian dead reckoning (PDR) system was field tested in different buildings. It yielded accurate matching results as well as a significant enhancement in positioning accuracy. Experimental results demonstrate that the mean absolute position error is less than 1.3 m and 95% confidence interval is between -3.16 m to 3.32 m. To further improvise the performance of map aided PDR system an extension to map based positioning is proposed via using landmarks. Landmark based positioning uses human as a sensor to sense proximity to landmarks. Landmarks are nothing specific as such but objects that are unique enough in comparison to the adjacent items e.g. quick response (QR) codes. Experimental results demonstrate that when map based positioning is used in addition to landmark based positioning the mean absolute position error is less than 1.0 m and 95% confidence interval is between -2.0 m to 2.0 m. Smartphones are mostly held in hands however these can be used as a lieu to dedicated wearable gadgets e.g. smart glasses that contain the similar set of sensors as smartphones. Hence, we investigate a scenario similar to smart glasses via smartphone mounted on helmet. The thesis concludes that in principle it is possible to logically localize a pedestrian within buildings using the inertial sensors embedded in smartphone. The algorithms developed in this thesis are suited to cases in which it is impossible or impractical to install large amounts of fixed infrastructure into the environment in advance. Also, methods proposed in this thesis are applicable in indoor tracking applications.