Speaker: Rafael Radkowski, Assistant Professor in the Department of Mechanical Engineering at Iowa State University
Title: Real-Time Vision-Based Object Recognition and Tracking for Augmented Reality Applications in Manufacturing
Abstract: Vision-based object recognition, detection, and tracking is important functionality for augmented reality applications in manufacturing; including quality control, training, and safety. Dr. Radkowski’s research aims on object recognition / tracking in point clouds with focus on real-time (camera frame time), robustness, and fidelity. Approaches under investigation are feature-based curvature descriptors and convolutional neuronal networks. A feature descriptor represents the curvature at a point on a surface as a set of axes and angles. Objects can be recognized by matching features of a point cloud with features of an object of interest. However, typical solutions are prone to outliers, which render them inapplicable in cluttered scenes and in the presence of users. We investigated an innovative descriptor matching solution that incorporates the topology of the object, which tremendously increases the detection (detection-by-alignment) robustness in cluttered scenes. Our method facilitates real-time tracking and achieves a registration fidelity up to 0.5cm using a low-fidelity camera. Recently, we investigate convolutional neuronal networks, focusing on network architectures and object representation for recognition and pose estimation using range data. Our goal is a re-usable network that only requires minimal re-training. The current objective is to determine an architecture – data set combination that yields high matching accuracies. We examined, for instance, the difference when using point range, normal vectors, point curvature, and combinations of them to identify (classify) objects. The combination of normal vectors and curvatures yield the highest accuracy, up to 80%. In his presentation, he will introduce his research goal by starting with an introduction of different applications, followed by an explanation of the two different approaches I focus on, as well as the recent research achievements in this area.
Bio: Rafael Radkowski is Assistant Professor in the Department of Mechanical Engineering at Iowa State University. He received his doctoral degree (with honors) from the University of Paderborn, Germany in 2006. After graduation, he became an acting leader of a research group in the Collaborative Research Center 614 at the Heinz Nixdorf Institute (Paderborn, Germany) and conducted his research for several years under the direction of Prof. Gausemeier. The research during this time focused on evolutionary algorithms and neuronal networks in application areas such as engineering design, virtual prototyping, and human computer interaction. He joined Iowa State University in 2012 as a post-doc and 2014 in his current position. Find his website here.