A FRAMEWORK FOR ACTIVELY SELECTING VIEWPOINTS IN OBJECT RECOGNITION

摘要: Object recognition problems in computer vision are often based on single image data processing. In various applications this processing can be extended to a complete sequence of images, usually received passively. contrast, we propose method for active object recognition, where camera is selectively moved around considered object. Doing so, aim at reliable classification results with clearly reduced amount necessary views by optimizing the movement access new viewpoints (viewpoint selection). Therefore, optimization criterion gain class discriminative information when observing appropriate next image. We show how apply an unsupervised reinforcement learning algorithm that problem. Specifically, focus modeling continuous states, actions and supporting rewards optimized recognition. also present sequential fusion gathered combine all these components into framework. The experimental evaluations split synthetic real objects one- or two-dimensional actions, respectively. This allows systematic evaluation theoretical correctness as well practical applicability proposed method. Our experiments showed combined viewpoint selection approach able significantly improve rates compared passive randomly chosen views.