Using stereo vision analysis to recognize partially obscured objects

Abstract

A method is proposed to recognize and locate the 3-D objects in the hidden case. Accurate camera calibration and the subpixel edge location algorithm is performed to get precise 3-D data. The algorithms for recognition and location contain three stages. First, 3-D data is analyzed to see how many objects are in the scene. In the second stage, the rotational and transitional parameters are determined only by using 3-D contour derived from stereo images to match to the object models. Finally, according to the matched object model and the rotational and translational parameters the visible line drawings are generated and used as a guide for searching correspondent line segments in original stereo image.<>