3D Object Tracking Using Analysis/Synthesis Techniques

Abstract

In post-production, a traditional method for creating some special effects is named “rotoscopy”. This technique consists of segmenting a video sequence by hand and for every frame. Our method is a new tool designed to reduce considerably the cost of this operation by making it almost automatic and quick. In our case, we track a rigid object whose geometry is known, in a sequence of video images. This new approach is based upon a two-steps process: first, one or several “keyframes” are used in a preliminary interactive calibration session, so that a 3D model of this object is positioned correctly on these images (its projection fits to the object in the image). We use this match to texture the 3D model with its image data. Then, a 3D predictor gives a position of the object model in the next image and the fine tuning of this position is obtained by simply minimizing the error between the textured model in this position and the real image of the object. Minimization is performed with respect to the 6 DOF (Degrees of Freedom) of the model position (3 translation parameters and 3 rotation ones). This procedure is iterated at each frame. Test sequences show how robust the method is.