Naslov
Model based inertial measuring of the kinematics of sit-to-stand movement

Autori
Musić, Josip

Vrsta, podvrsta i kategorija rada
Ocjenski radovi, magistarski rad

Fakultet
Fakultet za elektrotehniku

Mjesto
Ljubljana, Slovenija

Datum
05.11

Godina
2007

Stranica
131

Mentor
Doc. dr. Roman Kamnik

Ključne riječi
inertial sensors; sit-to-stand; dynamic human body model; extended Kalman filter; robot assisted standing up

Sažetak
The thesis deals with design and implementation of the system for recording human motion kinematics during sit-to-stand movement. The system is aimed to be used in robot assisted training of standing up. Robot assisted standing up training is used as a rehabilitation approach for re-learning of motion patterns in cases when subject had lost part of lower limb functionality. Traditionally the measurements of kinematics in standing up are performed by optoelectronic systems which provide accurate measurement but are expensive, require complex setting up procedure, have low portability and problem of marker occlusion making them unsuitable for practical ambulatory applications. This led to development of more practical kinematic measurement techniques, one of which is based on inertial sensors. Accelerometers and gyroscopes are part of inertial sensor's group and are characterized by low price, low power consumption, small size, light weight and high portability but also by some drawbacks which are mainly associated with their measurement accuracy. In the thesis an inertial sensors based motion tracking system is developed. The system incorporates five 3D inertial sensor units comprising accelerometers and gyroscopes and a dynamic human body model. Fusion of data from different sources is achieved by means of Extended Kalman filtering (EKF) approach. In the proposed method, the dynamic human body model is designed using principles of Lagrangian dynamics. During modelling the following assumptions were introduced: model consisted of three segments (shank, thigh and HAT) and three joints (ankle, knee and hip) and is valid only for time when there is no interaction between seat and human body. Also, segments are presumed to be rigid bodies with their masses contained at one point, center of mass (CoM) and are interconnected by ideal rotational joints with no added friction during rotation. Model movement is restricted to sagittal plane with assumption of symmetry of motion with respect to sagittal plane. In modelling three complex, non-linear, highly coupled equations were derived describing human motion in standing up. Model equations were implemented in Simulink simulation environment and its response compared to response of equivalent model constructed in Dymola-Modelica. Comparison of results verified the constructed model. Data from inertial sensors and data from dynamic human body model were fused using Kalman filtering. Due to non-linearities in the system the Extended Kalman filter was used. Computational efficiency was achieved by constructing the linear process equation enabling precalculation of system matrix A. Kalman filter state vector incorporated nine components pertaining to angles, angular velocities and angular accelerations of every segment. Measurement vector incorporated 12 components containing directly and indirectly measurable variables. Due to non-linearities present in measurement equation, calculation of Jacobian matrix in each step was necessary. Designed EKF structure was tested by the help of Simulink model in pendulum configuration. Comparison of estimated values with true model data showed good matching what verified the designed EKF structure. Experimental evaluation was accomplished by the help of measurement of healthy subject performing standing up maneuver. In several measurement tracks subject performed sit-to-stand movement with two different speeds: self-selected/normal and high speed. Obtained results were analyzed and compared to reference data measured by optical measurement system. Results show that the proposed method based on inertial sensors and three segment dynamic human body model demonstrated good matching in segment's orientation estimation with RMSE of 3.7o for shank, 3.8o for thigh and 5.2o for HAT segment. Performance in terms of angular velocity estimation is good while it is less accurate in case of angular acceleration. Obtained results demonstrated feasibility of the proposed method to be used as kinematic measurement system in robot assisted training of sit-to-stand motion. On the basis of results some improvements and additional testing were proposed.

Izvorni jezik
Engleski

Znanstvena područja
Elektrotehnika

Napomena
Komentor: Prof. dr. Vlasta Zanchi

POVEZANOST RADA