Research Of Body Movement Recognition System Based On Multi-sensor Data Fusion

With the rapid development of information industry and the raising popularization of personal computer, smart phone, intelligent appliance, and the other devices, HCI(Human Computer Interaction) plays a more and more important role in daily life. As one of the core technologies of the HCI field, body movement recognition is widely applied to medical treatment, education, security and protection monitoring, entertainment, and many other aspects.The main sources of movement recognition can be divided into image sensors and wearable sensors. The former can be affected by ambient light, object occlusion, and the other factors, and the latter is influenced by movement and sensor performance. According to the pros and cons of these two kinds, a movement data collecting system is designed and implemented based on multi-sensor, including inertial sensor, magnetic sensor and image sensor. Movement data are obtained and analyzed of different parts of human body, which will provide stable and robust support to recognition. The main studies of this paper is as follows.(1) The entire structure of the system is designed according to the analysis of body movement and practical requirement.(2) The multi-sensor data terminal is implemented, of which inertial sensor and magnetic sensor are used as the core. The works including hardware circuit design and commissioning, software programming of collection flow, data transmission of several nodes, the Euler angle algorithm, and data presentation on PC.(3) The image data terminal is developed based on depth image sensor, and the skeleton model of human body is built.(4) The magnetic data is analyzed and calibrated without an external reference, and the geomagnetic field is measured as standard of position.(5) Three data sets of daily actions were collected and analyzed by multi-sensor terminal.(6) Skeleton data under occluded conditions were analyzed and compensated by Euler angle.