My group started work in the then new area of Machine Learning In Computer Vision in 1994, and the focus has been the development of machine learning and statistical analysis techniques for solving problems within computer vision. Our projects have included learning of object models for use in model-based vision, high resolution feature extraction and recognition in satellite images and aerial photographs, and feature recognition and diagnosis in medical images.
In the area of Robotic Vision and Control, my group's focus is on developmental tools and environments for complex robot controller design, as well as active vision. Under tools and languages, we have used synchronous languages such as Esterel for controller design, simulation, implementation and testing. We have also developed a remotely controlled robot vision system that was later moved to the internet.
We have developed active vision techniques and applied them to robots, and this area is called Visual Robotics. Group members have developed an autonomous and adaptive security camera and a human motion classification system for it and robot navigation for a wheeled robot using active vision.
The techniques studied include process algebraic and synchronous models of concurrency and verification using simulation, refinement and temporal logic. We have begun looking at model-checking techniques at this point.
In the area of robot control design, I have designed a temporal-logic based verification technique for statecharts, which allows concurrent real-time properties tobe verified against a statecharts model of the robot system. I have extended these techniques to model and analyze real-time scheduling algorithms. practical design and development of robot control has been described under the robotics projects.
Since 1997, my group has been researching a series of related problems under adaptive design reuse for hardware (VLSI and SoC) and software designs. The main problem studied is the issue of how to take advantage of the large number of existing designs, in the form of soft core, components and program modules, and come up with better and faster new designs to solve new problems. A series of adaptive design reuse techniques have been developed by us and are applicable to designs written in CFSM (the description language used within POLIS, a tool for hardware/software codesign), , Argos, Esterel (both of which are synchronous progrmaming languages), symbolic Finite State Machines (that can handle control and data), and currently VHDL (a hardware description language).