Identifying movements in noisy crowd analytics data

Privacy-preserved tracking of WiFi-enabled devices such as smartphones offers a highly scalable solution for large-scale crowd movement studies. However, extracting knowledge out of pedestrian-tracking data acquired this way is not simple. This is, generally, due to the inherent inaccuracy of the measurement technique. Segmenting an individual's trajectory data into periods of stops and moves is a fundamental step in analyzing crowds' movement. Such distinctions allow us to answer advanced questions regarding visited locations or even social behavior. Algorithms previously designed for distinguishing movements from stay periods, assume datasets are gathered using GPS, which offers precise positioning. WiFi tracking, however, does not offer such precision. The location of devices can at best be reduced to a large area around the WiFi scanner. In this paper, we study a set of established algorithms for detecting periods of stops and moves from GPS-based datasets and their applicability to WiFi-based data. Consequently, we propose possible improvements to such algorithms considering the inherent characteristics of WiFi tracking data.