Speakers - Enrique Dunn



Enrique Dunn
Stevens Institute of Technology, USA

Title: Data-Driven 3D Modeling

Abstract: The visual inference of geometric and semantic concepts from capture imagery is still an open research problem being vigorously researched both in academia as in industry. The pervasive generation and public dissemination of visual media offer an ever-increasing datum of observed environments and events. The sheer amount of imagery and the wide-ranging diversity of recorded content renders the analysis of said data a challenging technical task. Large-scale visual 3D modeling provides a framework to both integrate such heterogeneous imagery into a common reference and synthesize rich environmental representations. In many regards, image-based geometry estimation has achieved a level of maturity that renders it as a “deployable technology” in consumer electronics (e.g. Microsoft’s Kinect Sensor) and enterprise-level data services (e.g. Google/Apple/Bing Maps). However, the continuous influx of video and image data (available in public archival and crowd-sourced repositories, social media, and live feeds) opens a diverse set of new challenges and opportunities for the deployment of 3D modeling systems. The first challenge (and perhaps the most evident) is achieving computational scalability in the presence of Internet-scale unstructured imagery datasets. Second, exploiting the heterogeneous nature of the available data to enhance (rather than hinder) the attained environmental representations. Third, modeling the observed environmental dynamics from uncontrolled imagery. In this talk, I will discuss recent research efforts aimed at augmenting the computational and conceptual scope of image-based 3D modeling in the context of crowd-sourced visual data. More specifically, I will present solutions to the aforementioned challenges by addressing the problems of large-scale data association, enhanced model fidelity through multi-source visual media integration, and spatiotemporal structure modeling from ad-hoc and unsynchronized video capture. The relevance of these technologies to fields such as remote sensing, robotics, virtual reality and human computer interaction will also be discussed.

BIO: Enrique Dunn is an Associate Professor in the Department of Computer Science of the Stevens Institute of Technology (Hoboken, New Jersey, USA). Prior to joining Stevens IT, Dr. Dunn was first a postdoctoral researcher, and subsequently, a Research Assistant Professor in the Department of Computer Science of the University of North Carolina at Chapel Hill. He holds a degree in computer engineering from the Autonomous University of Baja California (UABC Mexico), as well as a master’s degree in Computer Science and a doctorate in Electronics and Telecommunications, both from the Ensenada Center for Scientific Research and Higher Education (CICESE Mexico). His research on 3D Computer Vision focuses on geometric and semantic scene modeling in the context of large and heterogeneous visual data spanning multiple modalities. Dr. Dunn has authored over 45 papers in international conferences and journals within the fields of computer vision and photogrammetry. He is a regular reviewer for the top computer vision conferences (e.g. CVPR, ICCV, ECCV, SIGHGRAPH) and is a member of the Editorial Board of Elsevier Journal of Image and Vision Computing