M.Sc Thesis

M.Sc StudentSenesh Merav
SubjectSkin Movement Artifact Modeling and Compensation in Marker-
Based Human Motion Estimation for Biomechanics
DepartmentDepartment of Mechanical Engineering
Supervisor PROF. Alon Wolf
Full Thesis textFull thesis text - English Version


Quantitative analysis of human movement as an indication for bone and joint diseas is significant in both biomechanical and clinical studies. This analysis require estimation of three-dimensional position and orientation of bones during movement. Stero-photogrammetry measurements can reconstruct the trajectories of markers attached to the skin surface of the body segment that is being analyzed.  Theses trajectories are used to calculate the pose of the underlying bone, with the erroneous assumption that markers and bone segments are rigidly connected. It is well known that markers which are placed on the skin move with respect to the underlying bones, and this motion is task-dependent. Therefore, estimation of the skeletal motion from the observed skin attached markers is significantly affected by these soft tissue motions and may result in erroneous calculation of the joint kinematics parameters. Consequently, the soft tissue artifact was recognized as a major source of error in human motion analysis and is the primary factor limiting the resolution of detailed joint movement using skin-based measurement systems.  The majority of studies describing three dimensional in vivo segment motions do not account for errors associated with non-rigid body movement. Previous works approached the problem by modeling the segment as a rigid body and applying various estimation algorithms to obtain an optimal estimate of underlying skeletal motion, subject to a rigid body constraint. However, an accepted solution has not yet been developed.    

In this work we attempted to determine applicability of skeleton motion estimation based on imprecise skin-based measurements. The estimation was approached using both dynamical and statistical methods. The dynamic estimation method is based on the implementation of a Lagrangian approach to drive a model based procedure for the estimation of the rigid body motion form the measurements of markers attached to the skin. This approach takes into consideration the specific dynamic characteristics of the elastic and rigid component, which are different among subjects. The statistical estimation composes of Kalman filter with a numerical model and point cluster technique (PCT). Both methods were tested and evaluated using computerized and experimental dynamic models.

A good agreement between the estimated motion and the true measured motion was obtained using the dynamic method.

The use of filter in the statistical estimation method obtained better results than the use of PCT solely. However, Kalman filter found to have no preference over the use of a simple low pass filter.