Computer Vision and Robotics Laboratory
  George Mason University, Fairfax, Virginia

  Department of Computer Science
  MSN 4A5 Science and Technology II
  4400 University Drive, Fairfax, VA 22032
  Phone: (703) 993-1530, Fax:  (703)-993-1710

Computer Vision and Robotics Laboratory concentrates on the use of learning strategies to enhance visual performance
and development of new techniques for characterization of  geometric  and dynamic properties of the environments.
Research efforts focus on  the areas of human-computer intelligent interaction, biometrics, data compression and fractal image
representations, object recognition, motion analysis and stabilization, attention and control mechanisms, automatic target recognition,
3D  structure and motion recovery and intelligent agents for autonomous navigation.


Jim Chen
Zoran Duric
Jana Kosecka
Harry Wechsler

Affiliated Faculty

Peter Pachowicz

Research Activities

Human Computer Interaction
Pattern Recognition and Neural Networks
Biometrics and Forensic Laboratory
Mobile Robotics
Building Models from Video and Photographs

Students , Visitors, Alumni

  Related Links

   Computer Graphics Laboratory
   Center for Parallel and Distributed Computation
   Computer Science Department
   School of Information Technologes and Engineering (ITE)
   George Mason University


Human Computer Interaction

Activities concern Human - Computer Intelligent Interaction (HCII) and Human-Centered Systems (HCS). In every use of computers to solve human problems, a central and crucial factor is the flow of information and control between human and machine.  Present Human-Computer Interaction (HCI) technology constitutes a significant bottleneck for realizing the potential of computer technology.  Human-Computer Intelligent Interaction (HCII) goes beyond HCI as it involves organizing intelligent agents in a suitable framework, structuring the communication links between agents to suit task requirements, and presenting the information to the receiving agents in such a way as to utilize the particular agent's perceptual skills and to facilitate processing of the information. HCII leads also to a human-centered approach for the design of intelligent systems as it helps with creating a richer, more versatile, and effective virtual environment that supports human activity. Thus, the emphasis is not on just building autonomous systems that mimic humans but rather supporting human activity using intelligent system tools subject to the constraints, goals, and principles of human-centered (design and) systems (HCS).
           Relevant Courses : CS 777, INFT 835, CS682
         Faculty :   Jim Chen.   Zoran Duric,   Jana Kosecka,   Harry Wechsler 

Pattern Recognition and Neural Networks

Pattern Recognition encompasses a wide range of information processing problems of great practical significance, ranging from classification of handwritten or printed text to multimedia -- video and audio -- indexing and retrieval and data mining.  The approach we follow is hybrid and involves both statistical pattern recognition based on the probabilistic nature of the data  and machine learning techniques characteristic of AI. Traditional neural (connectionist) computation including feedforward and recurrent networks,  radial basis  functions, and self-organization feature maps,  statistical learning theory and support vector machines,  and mixture of experts are just some of the approaches being used.

          Relevant Courses : CS 688, CS 750, INFT 844
         Faculty:   Peter Pachowicz,   Harry Wechsler

Biometrics and Forensic Laboratory

The goal of this effort is to develop the science and technology of verifying a person's identity. Biometric measures the physical characteristics that make each of us unique, like the fingerprints, an eye's retina or iris, a face, a hand, a voice - and uses those measurements to confirm personal identity. Passwords are difficult to remember and easy to steal. Keys, driver's licenses and passports can be lost or forged. The human body, on the other hand, can't be forgotten, stolen, forged or misplaced. Practical uses for such biometrics are wide spread and include maintaining the security for both physical space and cyberspace. In particular, biometrics helps with controlling access to an office, a lab, or ATM, confirm the identity of buyers and sellers, protecting company networks from hackers, make electronic commerce safe and reliable, confirm student identity for distant learning and examinations, and keeping medical records on the web private. As face recognition technology requires little or no cooperation from the subject, it is becoming one of the top choices for biometrics and is starting to move into the commercial market. The biometrics and forensic group at GMU led by Professor Harry Wechsler has been actively engaged in face recognition research for the last 8 years under the FERET government sponsored R&D program. The R&D carried out at GMU involves face recognition, contents - based image retrieval, surveillance, ID verification, gender and ethnic classification, data compression, human studies, performance evaluation, and most recently video processing and interpretation of human activity.
Professor Wechsler has organized and directed for NATO an Advanced Study  Institute on Face Recognition : From Theory to Applications, held in Stirling, UK in 1997. The proceedings of the meeting have been published by Springer - Verlag in 1998. 20
Relevant Courses : CS 682, CS777, INFT 844,
Faculty :  Jim Chen,  Zoran Duric,  Harry Wechsler


Mobile Robotics - Vision and Action

The sophistication of the task mobile robots can achieve to a large extend depends of the representation of the environment they reside in. The central theme is to study visual sensing for robotic systems with the objective to advance the autonomy, reliability of their operation and ease of interaction of humans with robots. The complexity of the tasks robotics agents can achieve is directly linked to the representation of the environment they reside in. Visual sensing opens an avenue for estimating various properties from single or multiple images. Capturing properties of the environment some of which can be changing over time are crucial for successful interaction of the robotic system with the unknown and dynamically changing environment. We are currently investigating issues of representation of the environment in the context of mobile robot navigation, in indoors house/office environments, map building and vision-based control for various tracking and visual servoing tasks.
Relevant Courses : CS 685, CS682
Faculty :  Jana Kosecka, Zoran Duric


Building Models from Video and Photographs

 The focus of this research effort is to develop new techniques for building models of 3D environment from video or photographs. In case the knowledge of the camera and camera's motion is not available, the techniques for structure and motion estimation from image sequences are employed towards the recovery of these unknown quantities. The use of image based techniques for obtaining 3D models of the environment and/or objects can greatly enhance realism of the models as well as ease of obtaining them. These techniques cater to computer graphics, virtual reality and augmented reality applications, with the goal of obtaining greater realism. The research emphasis is on the algorithms for camera self-calibration, estimation of metric properties of the environment and techniques for model based motion and structure estimation which mediate the estimation of consistent representation of the environment. We are studying issues of acquisition of both large scale models and object level models, with the goal of capturing the essential 3D structure of the environment and selectively maintaining a collection of 2D views.
Relevant Courses : INFT 840
Faculty :  Jana Kosecka