Computer Vision Group
Computer Vision at the University of Maryland has a history of more than 40 years. The Computer Vision Laboratory was established by Prof. Azriel Rosenfeld in the 1960s. He is widely regarded as the father of the field, having written the first textbook on the subject, established the first scientific journal (Computer Graphics and Image Processing), and founded the first international conference (International Conference on Computer Vision). He published over 500 journal and conference papers on almost all aspects of computer vision and image processing.
The Laboratory has had a number of computer science alumni go on to great success around the world. For example, Narendra Ahuja at the University of Illinois is a Chaired Professor and a recipient of the prestigious IEEE Peori Prize; Chuck Dyer is a professor and former Chair of the top ten computer science department at the University of Wisconsin and Shmuel Peleg is a Professor and former Chair of Israels best computer science program at Hebrew University. They are all highlighted on the Departments web-based Alumni Hall of Fame. Over 100 computer science students have received their Ph.D.s in computer vision since the Laboratory was established.
Professor Yiannis Aloimonos began his career studying problems in video analysis, especially the problems of estimating the movement of a camera through space (think of a camera mounted on a robot vehicle) along with three dimensional models of the world through which the camera is moving (the so-called structure from motion problem). More recently, he has been studying models of human motor control and movement recognition that could be used by computer vision systems to model and recognize the movements that people make during everyday activities. Yiannis and his students have contributed to many other areas of computer vision, including stereo, camera design, camera calibration, and perceptual organization.
Professor Larry Davis, current Chair of the Computer Science Department, was a student of Prof. Rosenfelds in the 1970s. He has been on the faculty at Maryland since 1981. Early in his career he and his students studied problems related to object recognition based on shape as well as visual navigation systems for robot vehicles (especially automatic road following and more recently collision avoidance). For the past 15 years he has focused on visual surveillance systems. Here, the core computer vision problems involve methods to detect and track the motion of people and vehicles through collections of video cameras, logic-based methods for representing and recognizing activities involving people, vehicles and the objects they transport for detection of security and safety violations, and intelligent control of large surveillance camera networks. Currently, he is collaborating with researchers at Siemens Corporate Research to integrate some of these vision and analysis algorithms into a system that can identify unattended packages in public spaces such as airports, and with Honeywell Corporation on tracking people through very large camera networks.
David Jacobs is an Associate Professor in the department. He came to Maryland from NEC Research in Princeton. His primary research focus has been on the problem of how to recognize objects in images. This is a central problem in vision that lies between concrete, lower-level vision problems and central cognitive systems that attempt to tie semantics to the world. It is also a critical problem in many applications in multimedia and robotics. He has also done quite a bit of work in the computational and psychological study of perceptual organization. This is the lower level problem of how to identify the surfaces of potential objects before recognizing them. His recent work in recognition has mainly focused on understanding how to recognize objects under varying lighting conditions, and how to match images of objects that may deform or otherwise vary non-rigidly. He has also been interested in using techniques from computer vision in developing user interfaces and in graphics. His group is applying some of this work to build an electronic field guide for botanists, using object recognition methods to automatically match the leaves of plants, shown below. Ramani
Ramani Duraiswami is an Assistant Professor in the department who works at the interface of scientific computing, computational audio and computer vision. He is interested in efficient scientific computing algorithms and their application to fundamental problems in computer vision. Ramani, along with Nail Gumerov a research scientist in UMIACS have done a significant amount of work on fast multiple methods, of which the Fast Gauss Transform is an important instance. Ramani has applied this scientific computing research to vision problems including image segmentation, tracking and machine learning. These algorithms are becoming more important as large image and video databases can easily be collected. Ramani has an exciting research program on spatial audio creating true 360 degree sound landscapes for future music systems and virtual reality. He is also exploring the combined use of computer audition with vision.
Professor Hanan Samet has been with the department for thirty years and during that time has established himself as the worlds leading authority on spatial data representations and spatial database systems. His recently released monograph Foundations of Multidimensional and Metric Data Structures is already recognized as the definitive publication on methods for representing and searching spatial data. In his forward to the book, Turing Award winner Jim Gray of Microsoft wrote: Hanan Samet is the dean of spatial indexing... This book is encyclopedic... this book will be invaluable for those of us who struggle with spatial data, or scientific datasets, or with graphics, and vision problems involving volumetric queries, or with higher dimensional datasets common in data mining.
Rama Chellappa is a Professor in the Institute for Advanced Computer Studies and the Department of Electrical and Computer Engineering and is also an Affiliate Professor in the Computer Science Department. Prof. Chellappa came to Maryland almost fifteen years ago from the University of Southern California. Like Yiannis, he also did early pioneering work on motion estimation and structure from motion for video analysis. He has been working on problems related to human face recognition for more than a decade, and is widely recognized as one of the leading authorities on this subject worldwide. Some of his ongoing projects include developing new techniques for distributed smart camera networks through an integrated exploration of distributed algorithms, embedded architectures, and software synthesis techniques, new algorithms that can capture the structure of the human body from multiple video images of a person in motion, and PDA based remote sentry devices for monitoring scene acoustic and video through secure wireless network. The figure below illustrates some of Ramas recent work on human gait analysis, and determining if a person is carrying a heavy load based on the way they walk.
The success of the Vision Laboratory is also due, in large part, to the contributions of its Research Scientists, especially Daniel DeMenthon, David Doermann, Cornelia Fermuller, and Yaser Yacoob. Daniel, currently on an assignment to the National Science Foundation as Program Manager of its Computer Vision Program, is an expert on estimating the pose of three dimensional objects in two dimensional images, among other areas. David and Yaser are both alumni of the lab who have stayed to establish their own reputations in the field. David is Director of the Language and Media Processing Laboratory and is an international leader in the field of document image analysis. Cornelia does research on human and computer vision. She has worked on computation of image motion, three-dimensional motion estimation, reconstruction and shape estimation, and scene segmentation. Her recent work on how uncertainty in vision computations leads to geometric illusions is generally regarded as ground-breaking (see illustrative illusion to the left). Yaser has worked on problems related to human movement and appearance for over a decade. His first project involved development of a vision system that could recognize human facial expressions from video. A robot that Yaser developed with IBM that recognizes and mimics human expression is shown to the left. One of its most famous applications was to the astronaut, Dave, in the film 2001: A Space Odyssey, as part of David Storks (Ricoh Research) celebrated book comparing technology in the year 1999 (HALs birth year) to the predictions in the classic film. Yasers program successfully recognized that Dave was upset the same conclusion reached by HAL in one of the films most famous scenes!CS
