Multi-sensor Data Fusion For Human Gait Analysis

Human walking contains important physiology, kinematic and dynamic information. There are many application prospect of human gait analysis in real life, such as monitoring the patient's recovery progress in clinical practice, the control strategy of bionic robot, etc. This thesis estab-lished a wearable gait analysis platform based on MEMS sensor and body sensor network. The platform can be used to collect acceleration, angular velocity and the geomagnetic signals in the process of walking movement. Accurate gait parameters can be calculated through sensor data fusion algorithm and error correction process. This paper is composed on the basis of summa-rizing the existing research content at home and abroad, focusing on the following issues:how to reduce the orientation estimation error and increase the accuracy of gait phases partition; how to analyze the symmetry between dual feet and the gait stability; how to eliminate the sensor misalignment and binding position deviation among multiple sensors; And the applicability of the widely used zero velocity update algorithm in the field of gait analysis. The main research contents are as follows:1. This research adopted micro-electro-mechanical sensor and human body sensor network, a wearable gait analysis platform was designed and developed. A lot of work has been conduced on the sensor sensitivity calibration, bulky error elimination, etc. As for the influence of sensor position on gait analysis, which has always been ignored by many researchers, this paper pro-posed a sensor alignment method based on"boresighting", which can effectively eliminate the initial binding error, and it can be used for initial alignment of multiple sensors as well.2. This thesis focused on multi-sensor data fusion algorithms, including extended Kalman filter algorithm (EKF) which was used to fuse different types of sensor data, and zero veloci-ty update (ZUPT) algorithm used for speed and position error correction. A multiple threshold method with constraints was proposed to eliminate false gait phase detection due to sensor data fluctuation. The above algorithms lay the foundation for the precise calculation of gait parame-ters.3. In literature, most of the gait analysis researches are based on parameters from single foot. Which means there is a lack of study about the collaborative strategy of both feet. This paper presented the study of bipedal coordination by data fusion of gait parameters from dual feet. In addition, this paper provided a novel method based on dynamic time warping (DTW) algorithm, which can be used to evaluate biped gait symmetry, gait stability and consistency.4. In order to solve the inapplicability of the widely used ZUPT algorithm in the field of gait analysis, this paper proposed an extended zero velocity updating algorithm (E-ZUPT). Three sensor nodes were placed in different parts of the unilateral lower limb respectively. Lower limb kinematics constraint model can be established using Denavit-Hartenberg (D-H) convention, and lower limb movement signals were fused when stance phase was detected. The position error minimization problem can be disposed by solving the Euler-Lagrange equation, in this way, the usage of zero velocity update algorithm can be extended to other parts of lower limbs.