An Hidden Markov Model based Complex Walking Pattern Recognition Algorithm

The popularity of smartphone enables the capability of sensing the human activity, which can be used to provide various intelligent context-aware services. Most existing methods on human motion mode recognition assume that all sensors are mounted in a fixed position on users' body while walking. However, it is inconvenient for a user to mount his/her phone in a specific position. When a user holds his/her phone in hand, the situation becomes fairly complex. First, the motion of the hand is coupled with the general activity of the user. Second, the characteristics of the inertial sensors may vary along with diverse carrying modes. In this paper, eight different human activities are defined to characterize the phone holding modes and the motion patterns. By extracting features in time and frequency domains from the tri-axis accelerometer and tri-axis gyroscope signals, we design and implement a hierarchical classification system to detect complex walking patterns based on the decision tree, random forest and hidden Markov model (HMM). Simulation experimental results demonstrate that the recognition success of complex walking pattern using the proposed method is more than 93.8% for eight complex motion modes.