A Research Of Human Body Attitude Recognition Based On Multi-sensor Information Fusion

In recent years, with the development of the micro-electro mechanical systems(MEMS) and the wireless body area network(WBAN), the wearable health care system has emerged. As a branch of the health care system, the human body attitude recognition system based on wireless body area network has become a new hotspot. However in previous studies, the only using of the acceleration sensor has led to the problems of single signal and poor robustness of the motion monitor system. To solve this problem, researchers have proposed to increase the number of sensor nodes and design complete sensor fusion algorithms, But the drawbacks that come along might be worse portability, less wearable and more complicated recognition algorithm.Based on the above situations, this paper do research in the human body attitude recognition based on multi-sensor fusion. Firstly, a human routine motion data Acquisition platform is established which is based on WBAN, aiming to collect, transmit, display and storage the human motion information in real-time. Then the human body attitude angle estimation algorithm is designed which is based on PI regulation and complementary filter, estimate the attitude angle in time. Finally, attitude angle、acceleration characteristics and altitude the attitude are utilized as latent parameters to build the recognition algorithm, which initially realize the function of human routine attitude recognition.The main work of this paper is concluded as follows:(1) On the basis of existing inertial measurement module, this paper design a portable human motion sensor terminal, which consists of sensing devices (including a three-axis accelerometer, a three-axis magnetometer and a three-axis gyroscope), WIFI module, micro-controller module, SD storage module. This sensor node can provide real-time information (such as acceleration, velocity and magnetic field strength) during human movement to the upper computer continuously. A PC side monitoring center software based on VB application framework is exploited, which can receive and display data of sensor nodes in real-time, provide initial configuration of the sensor (sensor calibration, etc.), and simple database management functions, such as data archiving and query. (2) In order to the problems of poor stability and short precision in traditional attitude estimation algorithms, a method based on multi-sensor information fusion is proposed. The quaternion is used to calculate the attitude changes, and deviations between measurements and observations from the Inertial Measurement Unit are managed by PI regulator. The complementary filtering is adopted for multi-sensor information fusion, and the real-time attitude is got, and can output the high precision attitude steadily.(3) The motion data is acquired by the human motion experiment, and it is used to estimate the body attitude angle and extract features. Then the attitude is roughly classified roughly by using classification algorithms to improve the rapidity of body attitude recognition. Finally a decision tree attitude recognition algorithm is designed which combine with the rough classification, which can realize the human daily motion accurate recognition.