Matthias Wimmer, Christoph Mayer, Freek Stulp, and Bernd Radig. Estimating Natural Activity by Fitting 3D Models via Learned Objective Functions. In Workshop on Vision, Modeling, and Visualization (VMV), pages 233–241, Saarbrücken, Germany, November 2007.
Model-based image interpretation has proven to robustly extract high-level scene descriptors from raw image data. Furthermore, geometric texture models represent a fundamental component for visualizing real-world scenarios. However, the motion of the model and the real-world object must be similar in order to portray natural activity. Again, this information can be determined by inspecting images via model-based image interpretation. This paper sketches the challenge of fitting models to images, describes the shortcomings of current approaches and proposes a technique based on machine learning techniques. We identify the objective function as a crucial component for fitting models to images. Furthermore, we state preferable properties of these functions and we propose to learn such a function from manually annotated example images.