Doctoral Defense by Loren Schwarz

Date: Friday June 22nd, 2012 Time: 14:00 Location: Room 00.05.035, FMI-Building, Garching, Boltzmannstr. 3

Machine Learning for Human Motion Analysis and Gesture Recognition

Abstract:

In this thesis, we investigate machine learning methods for human motion analysis. We introduce algorithms for human pose estimation and activity recognition that do not build upon conventional cameras and that can cope with noisy and incomplete input data. We propose methods that capture human movements using body-worn inertial sensors or using a depth camera. In a training phase, the measurements from these modalities are complemented with precise and complete movement information recorded with a camera-based motion capture system. A novel approach for learning models of human motion is introduced based on non-linear dimensionality reduction and regression that enables human pose estimation and simultaneous activity recognition, given only the inertial sensor or depth data. As an application scenario, we present a customizable method for medical gesture-based human-machine interaction that fills an existing gap in the operating room, as confirmed by surgeons.