Sports Behavior Recognition Algorithms Based On Body Area Networks

With the improvement of living standard and the attention to health, more and more people pay attention to sports. Meanwhile, a large number of researchers are attracted to study sports. With the fast development of sensor network, the study of sports by using Body Area Networks has been a hot area in recent years. There are many applications of Body Area Networks in sports field, such as the recognition of human daily sports activities, the detection of martial motion sequences, the automatic video annotation and the automatic generation of highlights, performance sports, etc. Badminton is one of the most popular sports in the world and many people play badminton to keep healthy. A system based on Body Area Networks is designed in this paper, and the system not only can recognize12different badminton stroke activities but also can assess the skill of badminton athletes in3different levels.In this paper,4wireless inertial sensors are placed on right wrist, left wrist, right side of waist and right ankle to collect motion data of12athletes. The motion data is transmitted to terminal devices by wireless receiving node. A two-layer classification algorithm is proposed to recognize12different badminton stroke activities. In the first layer, acceleration magnitude of right wrist is used to determine a threshold to detect badminton strokes and HMM is used to filter non-stroke activities. In the second layer, HMM is adopted to classify all the candidate stroke activities into12categories. Experiment results show that average recognition accuracy of HMM to recognize12badminton stroke activities achieves99.3%, which is higher than other4classification algorithms. The99.3%recognition accuracy verifies the effectiveness and feasibility of the two-layer classification algorithm proposed in this paper. In addition, performance of smash activity of badminton athletes in3different levels are investigated. HMM algorithm is adopted to classify smash activities into3categories. The average recognition accuracy achieves100%, which shows that it is possible to design a badminton training system based on Body Area Networks. The ultimate purpose of the study is to design a badminton training system, which not only can recognize different stroke activities, but also can assess the skill of athletes and provide advice and feedback for athletes and coaches.