A WiFi Assisted Pedestrian Heading Estimation Method Using Gyroscope

In order to improve the performance of the indoor localization system, the fusion of multi-source data is a common approach. For example, one can improve the WiFi localization accuracy on smartphones by combining pedestrian dead reckoning (PDR) results obtained through inertial sensors embedded in smartphones. Though obvious improvement in localization can be achieved, the existing methods do not sufficiently exploit the advantages of two data sources. To be specific, the existing studies directly fuse WiFi localization results and PDR results at a high level, i.e., the final coordinates of the WiFi localization system integrate with the final coordinates of PDR by certain algorithms, but ignores their relationship at a low level, i.e, the heading of the PDR , not its location results, is improved by the help of the WiFi localization. In addition, it is acknowledged that the pedestrian heading is the major source determining the performance of PDR. Therefore, this paper proposes to design a novel pedestrian heading estimation by fusing PDR and WiFi at a low level. Different from the traditional method, which employs a magnetometer to eliminate the drifts of a gyroscope, the method utilizes only the gyroscope of a smartphone for the heading estimation and relies on the WiFi localization trajectory in the fusion to compensate for the drift errors of the gyroscope-based heading estimation. In our algorithm, firstly, a pedestrian's activities trajectory is segmented into several straight paths with the help of the gyroscope of a smartphone. Secondly, the WiFi fingerprint localization coordinates falling into the time window of each straight path are fitted by the least-squares linear regression method. Lastly, the deviations of the gyroscope heading estimation of the smartphone when pedestrians walk in a straight direction are mitigated using the fitting slope obtained by the WiFi localization. Extensive experimental results demonstrate that our proposed algorithm can efficiently estimate the heading of pedestrians, and effectively reduce the cumulative errors of the gyroscope-based heading estimation using smartphones. In our experiments, the average error of the heading for pedestrians in 294 steps was reduced from 24.3 degrees to 1.22 degrees. Not requiring a magnetometer, our algorithm can reduce the drift errors of the heading estimation of pedestrians, achieve the deeper fusion of multi-source data in the fusion of WiFi and PDR, and potentially improves the endurance of smartphones.