Hong Zhang, PhD

Email

hzhang@ualberta.ca

Education

• B.S., Electrical Engineering, Northeastern University, 1982
• Ph.D., Electrical Engineering, Purdue University, 1986

Areas

Robotics
Computer Vision and Multimedia Communications

Interests

Robotics, Computer Vision, Image Processing.

Summary

• Visual Robot Navigation
  We are interested in the general question of how a mobile robot navigates with the help of a camera. One approach we follow is that of appearance SLAM (simultaneous localization and mapping) in which a robot environment is described topologically in terms of images sampled at key locations (key frames). In our recent research, we have also compared the performance of visual feature extractors for visual SLAM. Most recently, we are investigating the use of efficient data structures (e.g., locality sensitive hashing and kNN graph) for matching visual features in large maps (with millions of features). In addition, we are developing algorithms in qualitative visual homing in order for a robot to navigate in its appearance map.
• Image Processing
  In image processing, our research focuses on image segmentation, and it is driven largely by the need of Alberta's oil sand mining industry to measure ore size while the oil sand ore is crushed, conveyed and screened. One novel image segmentation algorithm we have developed formulates image segmentation as one of pixel classification, which is then solved by machine learning. We also extensively exploit the known shape of the objects for their segmentation. To evaluate our segmentation algorithm objectively, we have designed a performance metric for images of multiple objects that fairly penalizes over- and under-segmentation. Our segmentation algorithms have been successfully deployed in practical applications.

• Collective Robotics
  In collective robotics, we are interested in understanding the underlying principles that enable multiple robots to work cooperatively in accomplishing joint tasks. Our approaches are biologically inspired in which behaviors of social insects are mapped to local rules of interaction among the robots. We have investigated general methodologies with which one can design collective robot systems, synthesize the rules of interaction, and prove about their properties. In recent years, we have focused on the tasks of collective decision making and collective construction, to ground our research ideas. Shown below are snapshots of collective construction via the bull dozing behavior (left), collective decision making (middle), and highlights of a Robocup match by Team Canuck based in Computing Science in 2000-2005 (right).