Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
405227 | Knowledge-Based Systems | 2013 | 11 Pages |
One of the key applications of Smart Environment (which is deployed with anonymous binary motion sensors [1] and [2]) is user activity behavior analysis. The necessary prerequisite to finding behavior knowledge of users is to mine trajectories from the massive amount of sensor data. However, it becomes more challenging when the Smart Environment has to use only non-invasive and binary sensing because of user privacy protection. Furthermore, the existing trajectory tracking algorithms mainly deal with tracking object either using sophisticated invasive and expensive sensors [3] and [4], or treating tracking as a Hidden Markov Model (HMM) which needs adequate training data set to obtain model’s parameter [5]. So, it is imperative to propose a framework which can distinguish different trajectories only based on collected data from anonymous binary motion sensors. In this paper, we propose a framework – Mining Trajectory from Anonymous Binary Motion Sensor Data (MiningTraMo) – that can mine valuable and trust-worthy motion trajectories from the massive amount of sensor data. The proposed solution makes use of both temporal and spatial information to remove the system noise and ambiguity caused by motion crossover and overlapping. Meanwhile, MiningTraMo introduces Multiple Pairs Best Trajectory Problem (MPBT), which is inspired by the multiple pairs shortest path algorithm in [6], to search the most possible trajectory using walking speed variance when there are several trajectory candidates. The time complexity of the proposed algorithms are analyzed and the accuracy performance is evaluated by some designed experiments which not only have ground truth, but also are the typical situation for real application. The mining experiment using real history data from a smart workspace is also finished to find the user’s behavior pattern.