Michael Greenspan was awarded a BSc ('86, Physics and Applied Mathematics) from the University of Toronto, a BASc ('89, Electrical Engineering) and MASc ('91, Electrical Engineering} from the University of Ottawa, and a PhD ('91, Systems and Computer Engineering) from Carleton University, Ottawa. From 1991 to 2001, Dr. Greenspan worked at the Institute for Information Technology of the National Research Council of Canada, as a researcher and ultimately as the leader of the Computational Video Group. He is currently a Professor with the Department of Electrical and Computer Engineering at Queen's University.
Michael has published over 30 technical papers and holds 3 patents. He has worked with industry on a number of collaborative projects. Michael is a member of the IEEE Computer Society, and holds an appointment on the Research Management Committee of Precarn Associates, Inc.
Research Interests
Object recognition and pose determination:
My main research interest is object recognition and pose determination, where the goal is to develop methods to automatically identify and localize known objects within image data. The challenge is to be both efficient and reliable. I am working with 3D range data, and have developed an object recognition method based upon geometric probing. This is a minimalist and (I believe) elegant technique which turns out to be a kind of efficient template matching in the discrete domain. We are currently applying these techniques to the problem of localizing and tracking the motion of a free-flying satellite for autonomous robotic capture.
Motion planning
Motion planning systems calculate robot movements which avoid colliding with any known or sensed obstacles in the robot's workspace. The motion planning problem can be partitioned into two sub problems: collision detection, and search. I have developed a representation for collision detection that is very efficient, executing well within real-time constraints. We are currently investigating the parallelization of enumerative search methods for the efficient solution manipulator motion planning problems.
Applied computational geometry
I have also developed TINN, which is a novel solution to the Nearest Neighbor Problem. Succinctly stated, the Nearest Neighbor Problem is: given a set P of n points, find the closest point in P to a query point q. The exhaustive way of solving the problem is to simply visit each point in P and calculate its distance to q, which costs n distance computations. Any efficient method of solving this problem must therefore identify the nearest neighbor in less than n distance computations. But beware - the cost of doing so must not exceed the cost of the computations that it is pruning! We are currently applying a variation of this method to accelerate the Iterative Closest Point Algorithm (ICP) for the registration of 2 point sets.
For more info on Dr. Greenspan's research visit RCV lab webpage
To view Dr. Greenspan's publications, please visit RCV lab webpage