Human 3D models and motion analysis are nowadays used in a wide range of applications, spanning from medicine to security and surveillance. In this work, we will focus on the creation of biomechanical models for clinical and sports application of these methods. State of the art motion capture systems are capable of measuring with sufficient accuracy the 3D coordinates of reflective markers place above the skin. The main issues clinical gait analysis is currently facing are the repeatability of the measurements and the compensation of soft-tissue motion. Here we present a general framework to automatically recover joint parameters modelling the human articulations, from the 3D coordinates of a point cloud provided by motion capture systems. Additionally, we describe an approach capable of recovering a more accurate rigid body description of non-rigid bodies, and able to deal with the problem of marker occlusion. We then propose a new quadratic model to explain soft-tissue artifacts formed as a natural extension of the existing rigid body models. The parameters of this model are computed using an algorithm for large scale non-linear optimisation. Finally we use synthetic data to assess the performance of the algorithms and to compare the results with ground truth data. Qualitative analysis of real data sequences over different motion capture databases are also presented. 1.

ABSTRACT: Human 3D models and motion analysis are nowadays used in a wide range of applications, spanning from medicine to security and surveillance. In this work, we will focus on the creation of biomechanical models for clinical and sports analysis. Nowadays, motion capture systems accurately measure the 3D coordinates of reflective markers place above the skin. Here we present a general framework to automatically recover joint parameters modelling the human articulations from the 3D coordinates of a point cloud provided by motion capture systems. Additionally, we describe an approach capable of recovering a more accurate rigid body description of non-rigid bodies. We then propose a new quadratic model to explain soft-tissue artefacts formed as a natural extension of the existing rigid body models. Finally we use synthetic data to assess the performance of the algorithms and to compare the results with ground truth data. Qualitative analyses of real data sequences over different motion capture databases are also presented. 1

This thesis proposal addresses the 3D reconstruction from feature correspondences among many images. An introduction to the reconstruction problem is given at the beginning. Then, the State-of-the-Art in the field of 3D reconstruction is presented. Our previous research related to this field is described. Finally, the thesis goals are formulated.

In range-only Simultaneous Localization and Mapping (SLAM), we are given a sequence of range measurements from a robot to fixed landmarks. We then attempt to simultaneously estimate the