A New Robust Motion Reconstruction Method based on Optimization with Redundant Constraints and Natural Coordinates.

Abstract

The three-dimensional analysis of human movement is of interest in many different fields of life sciences, computer animation and engineering. The elements involved in the analysis of human movement are usually measurement equipments for estimating kinematic, kinetic and myoelectric variables, mathematical models of the human musculoskeletal system, and mathematical methods for calculating the variables which cannot be directly measured. The aim of this thesis is to advance in the knowledge of four aspects of the three-dimensional analysis of the human movement: 1) the motion reconstruction of human movements using large and medium-size skeletal models with open- and closed-loops, 2) two problems inherent to optoelectronic motion capture systems: the missing marker problem and the impossibility of measuring completely the motion of some bones which move under the skin, 3) the estimation of subject-specific parameters using only a motion capture system, and 4) the development of several human skeletal models suitable for analysing different vehicle-related motions. The motion reconstruction problem using human skeletal models defined with natural coordinates is formulated as a nonlinear constrained optimisation problem with equality constraint equations. The main contribution of this thesis is a new optimisation method for solving the motion reconstruction problem. The new optimisation method can reconstruct the motion of large-size human skeletal models with open- and closed-loops defined with natural coordinates and it can also handle redundant constraints. Four new strategies have been proposed for solving the two problems inherent to optoelectronic motion capture systems addressed in this thesis. The four strategies have been evaluated using experimental motion data with satisfactory results. These strategies enable a more robust reconstruction of the human movement. The subject-specific parameters are estimated using methods based on the measurement of anatomical landmarks. Furthermore, a new measurement protocol for measuring the anatomical landmarks and a new methodology for estimating all subject-specific parameters from the measured anatomical landmarks are proposed. Three human skeletal models have been developed for studying driving manoeuvres: one upper body model with a detailed model of the shoulder complex and another upper body model with a simplified model of the shoulder complex for studying steering manoeuvres and one right lower limb model for studying braking manoeuvres. Additionally, a human skeletal model of the whole body, based on the RAMSIS model, has been used to study generic arm reaching motions and three types of vehicle-related motions.