GPS-based Noise Reduction Approach for Trajectory Recognition

Most people habitually follow certain routes for traveling to specific places. These personal routes have been investigated in order to recognize users' trajectories on the basis of the GPS coordinates of their mobile devices, such as smartphones. However, GPS involves an error of around 50 m. In particular, when users remain stationary or move in a building or tunnel, their location may be incorrectly represented by GPS in some cases. In this paper, we propose a method to detect incorrect GPS coordinates so that they can be disregarded while recognizing users' trajectories on the basis of the GPS coordinates of their smartphones. In addition, we describe three approaches: 2-filter-based noise reduction adapted from previous studies, noise-cell-based noise reduction, and 3-filter-based noise reduction using WiFi vector similarity as well as 2 filters. Experimental results show that when users move around a certain area (e.g., a university or office) that has several WiFi access points, the 3-filter-based approach using a WiFi filter has a better noise reduction ratio than other methods. On the other hand, it is observed that when moving around a city, the noise-cell-based approach has a better noise reduction ratio.