Conventional marker-based optical motion capture methods rely on scene attenuation (e.g. by infrared-pass filtering). This renders the images useless for development and testing of machine vision methods under natural conditions. Unfortunately, combining, calibrating and synchronising a system for motion capture with a separate camera is a costly and cumbersome task. To overcome this problem, we present a framework for efficient, omnidirectional head- pose initialisation and tracking in the presence of missing and false positive marker detections. As such, it finally enables easy, accurate and synchronous head-motion capture as ground truth with or input for other machine vision algorithms.
展开▼