A kinematic human upper body model for evaluating clothing

Abstract

In this thesis a kinematic human upper body model for evaluating clothing fit and appearance is developed and validated. Realistic and accurate human body models are required in many different application areas, including medicine, computer graphics, biomechanics, and sport science. A particular application of interest for a human body model is a virtual reality clothing model to evaluate fit and appearance of garments. A robotics-based model for the human upper body skeleton is derived. To validate the model, upper body motion data is collected with a markerless motion capture system using Microsoft Kinect. A baseline evaluation of markerless motion capture with a single Kinect sensor presents results from tracking a robot arm trajectory, frequency tests, and human motion capture experiments. Because occlusion causes a single Kinect sensor to fail in accurately predicting the human pose, a second Kinect sensor is integrated into the system. Data from the two sensors is fused and filtered using an Extended Kalman filter. The results are compared to marker-based tracking with a Vicon Motion Capture system. The Extended Kalman filter is shown to ensure constant body segment lengths, thus producing a more realistic estimation of the joint positions than obtained from the raw Kinect data. The proposed setup offers a low-cost, markerless, and portable alternative to marker based motion tracking.