Marker-less motion capture in general scenes with sparse multi-camera setups

Human motion-capture from videos is one of the fundamental problems in computer vision and computer graphics. Its applications can be found in a wide range of industries. Even with all the developments in the past years, industry and academia alike still rely on complex and expensive marker-based systems. Many state-of-the-art marker-less motioncapture methods come close to the performance of marker-based algorithms, but only when recording in highly controlled studio environments with exactly synchronized, static and sufficiently many cameras. While relative to marker-based systems, this yields an easier apparatus with a reduced setup time, the hurdles towards practical application are still large and the costs are considerable. By being constrained to a controlled studio, marker-less methods fail to fully play out their advantage of being able to capture scenes without actively modifying them. In the area of marker-less human motion-capture, this thesis proposes several novel algorithms for simplifying the motion-capture to be applicable in new general outdoor scenes. The first is an optical multi-video synchronization method which achieves subframe accuracy in general scenes. In this step, the synchronization parameters of multiple videos are estimated. Then, we propose a spatio-temporal motion-capture method which uses the synchronization parameters for accurate motion-capture with unsynchronized cameras. Afterwards, we propose a motion capture method that works with moving cameras, where multiple people are tracked even in front of cluttered and dynamic backgrounds with potentially moving cameras. Finally, we reduce the number of cameras employed by proposing a novel motion-capture method which uses as few as two cameras to capture high-quality motion in general environments, even outdoors. The methods proposed in this thesis can be adopted in many practical applications to achieve similar performance as complex motion-capture studios with a few consumer-grade cameras, such as mobile phones or GoPros, even for uncontrolled outdoor scenes.