Research And Implementation Of Users'motion Mode Detection Technology For The Intelligent Terminals

In recent years, with the rapid development of the mobile Internet and wireless sensor technology, various services based on intelligent terminals become an important part of people's daily life. Among them,intelligent terminal-based users' motion mode detection service is a basic service, which is a hot research field of current mobile application manufacturers and research institutions. Common users' motion modes include walking, going upstairs, going downstairs, running, jumping and so on. Most of the users' motion mode detection technology based on intelligent terminal collect the original data related to the users' motion mode, and then process and analyze the data using appropriate techniques and methods to sense users' motion modes.Users' motion mode detection technology is widely used in smart home, health care, sports training and other areas, so the study of this technology has important practical significance. In addition, users'motion mode detection technology is a basic research and can provide scientific research with powerful technical support, including users'motion mode contextual services.At this stage, most inertial sensor based users' motion mode recognition methods assume that sensors are mounted in a fixed position on users' body, such as users' feet or belt. When these sensors are fixed on the users' feet, the stance phases of the feet can be easily determined and periodic Zero velocity UPdaTes(ZUPTs) are performed to constrain position errors. However, it is inconvenient to mount a sensor on a specific position of users for a long time. To recognize users' motion modes more naturally, researchers tend to use smartphones to detect users'motion mode.In this dissertation, ten different motion modes are defined to characterize the phones' holding modes and the motion mode 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. We leverage decision tree, random forest to detect coarse-grained motion modes and fine-grained motion modes respectively. In the stage of detect user's motion modes through random forest, we decompose the acceleration into vertical component and horizontal component and extract features respectively to increase the dimension of feature extraction, so as to improve the detection accuracy.Finally, a hidden Markov model (HMM) is adopted to filter the noise and random errors in the detection process, which further improves the detection accuracy of motion mode.Experimental results demonstrate that the detection accuracy of the ten complex motion modes using the proposed method is more than 93.8%. The method has considerable feasibility and practicability which can provide guidance for other intelligent applications, indoor positioning technology and vehicle detection technology.