Model-Based 3D Object Recognition in RGB-D Images

A computational framework for 3D object recognition in RGB-D images is presented. The focus is on computer vision applications in indoor autonomous robotics, where objects need to be recognized either for the purpose of being grasped and manipulated by the robot, or where the entire scene must be recognized to allow high-level cognitive tasks to be performed. The framework integrates solutions for generic (i.e. type-based) object representation (e.g. semantic networks), trainable transformations between abstraction levels (e.g. by neural networks), reasoning under uncertain and partial data (e.g. Dynamic Bayesian Networks, Fuzzy Logic), optimized model-to-data matching (e.g. constraint optimization problems) and efficient search strategies (switching between data- and model-driven inference steps). The computational implementation of the object model and the object recognition strategy is presented in more details. Testing scenarios deal with the recognition of cups and bottles or household furniture. Conducted experiments and the chosen applications confirmed, that this approach is valid and may easily be adapted to multiple scenarios.