Movement behavior recognition based on statistical mobility sensing

Abstract

Movement behavior recognition is an important technology submitting to personalized location-based service (LBS), which includes such things as health monitoring, recruitment flow of information, logical localization and neighbor discovery. Based on the balance between energy consumption and recognition accuracy, we propose a kind of statistical sensing method for low-power movement behavior recognition by means of a smartphone based accelerometer. First, we analyze the temporal statistical characterization based on the sensory samples for both the moving behavior and the stationary one. Second, we use Bayes’ theorem to distinguish between these moving and stationary behaviors, further to detach the stationary behavior identification based on the standard deviation of their accelerations. Finally, moving behaviors (walking, running, cycling or motoring), can be identified in terms of combining the mobility with their statistical characterization amongst various behaviors. The experiment data was collected by 16 volunteers and the results show that our method not only achieves a 94% behavior classification accuracy rate, but also reduces energy consumption by 68.9% over DT-DHMM.