SCS Undergraduate Thesis Topics
|Matt Carson||Dave Touretzky||Blocks World Vision for the AIBO Robot|
Robust visual object recognition is still a largely unsolved problem. On the AIBO robot, color segmentation is robust,so object recognition in the Tekkotsu visual framework is based on mono-chromatic objects whose appearance does not change with orientation. This research seeks to expand upon existing object recognition tools by developing ways to detect objects with more complex features, multiple colors, and appearances that change based on orientation.
The research focuses on developing algorithms to identify brick-shaped objects in the image seen by the robot. One approach recognizes the edge features that are invariant in a brick object. Line extraction algorithms are developed and applied to the image, and possible bricks are then extracted from the lines in the image based on relative size and orientation of sets of lines. In a differing approach, knowledge of the colors of the faces of bricks is used to find adjacent blobs of color that could be brick faces. The corners of these blobs are extracted to find more precise information about the size and shape of the brick. These algorithms are tested on many different brick objects, and their applicability toward other regular shapes (cylinders and pyramids) is considered.