Feature extraction for video rate three dimensional imaging via coloured spots

Video-rate three dimensional (3D) acquisition systems are required, in particular for dynamic shape sensing of the mouth to reinforce vocal tract models of speech production. In a new structured light technique for 3D sensing, scenes are illuminated by an array of circular spots which are colour encoded to resolve spatial ambiguity. The projected spots are imaged as ellipses, their position and shape depends on the position and orientation of the illuminated 3D surface. This enables the formulation of a novel 3D Hough transform (HT) specifically for detection of the surface position and orientation via the imaged spots. In the HT accumulator array voting constraints are applied to minimise both the number or background votes cast and the likelihood of raise peaks. This new technique is demonstrated to extract a moving face successfully, as part of a video rate 3D acquisition system.