Abstract: We present a method for animating 3D models of animals from existing live video sequences such as wild life documentaries. Videos are first segmented into binary images on which Principal Component Analysis (PCA) is applied. The time-varying coordinates of the images in the PCA space are then used to generate 3D animation. This is done through interpolation with Radial Basis Functions (RBF) of 3D pose examples associated with a small set of key-images extracted from the video. In addition to this processing pipeline, our main contributions are: an automatic method for selecting the best set of key-images for which the designer will need to provide 3D pose examples. This method saves user time and effort since there is no more need for manual selection within the video and then trials and errors in the choice of key-images and 3D pose examples. As another contribution, we propose a simple algorithm based on PCA images to resolve 3D pose prediction ambiguities. These ambiguities are inherent to many animal gaits when only monocular view is available. The method is first evaluated on sequences of synthetic images of animal gaits, for which full 3D data is available. We achieve a good quality reconstruction of the input 3D motion from a single video sequence of its 2D rendering. We then illustrate the method by reconstructing animal gaits from live video of wild life documentaries.