الفهرس | Only 14 pages are availabe for public view |
Abstract Mobile robots require vision capabilities to sense and adapt to the environment. The task of 3-D object recognition is an important component of visual sensing. 3-D object recognition is however a very difficult task because of the infinite amount of variations in the 3-D real-life scene. In this thesis, an appearance-based 3-D object recognition model that recognizes 3-D objects from multiple 2-D views is proposed. The framework of the recognition model contains two main processes: Features extraction process and Learning and classification process. Three types of features are used, each independently, in the recognition process. First, Hu moment features are used for recognition, as they are invariant to translation, rotation and translation of objects. This property makes them suitable for the 3-D object recognition task. Second, a combination of the Hu invariants and the Affine Moment Invariants (AMIs), Hu-AMIs features, are used. AMIs are invariant to affine transformation and using them with Hu invariants improves the recognition of the model. Finally, the color features are used for recognition as they contain very useful information about the appearance of the objects. Learning and recognition are performed using the Support Vector Machines (SVMs) network. SVMs are newly introduced networks that have high generalization capability. Many experiments are performed to demonstrate the performance of each feature type together with the SVMs in recognition. The main experiments are: • Testing the performance of the recognition process as varying the number of training views per an |