In man-made environments, rectangular shapes can be seen almost everywhere. It is thus convenient to use rectangular shapes for robot location, which is equivalent to the problem of close-range space resection encountered in photogrammetric engineering. A new method for robot location using mutually parallel or perpendicular line pairs in observed rectangular shape images is proposed. From a monocular image of any object with at least a rectangular
shape, image processing and numerical analysis techniques are applied to extract the projection characteristics of each corner point in the rectangular shape. The position and orientation parameters of a camera mounted on the robot can then be computed for robot location. Multiple rectangular shapes can be utilized to promote the accuracy of the parameters. Partial rectangular shapes also can be utilized. A major merit of the method is that the solution can be uniquely and analytically determined without iterative computation. Experimental results show the feasibility of the proposed approach. Error analysis useful for determining location precision is also included.
Relation:
Photogrammetric Engineering and Remote Sensing 56 (2): 231-238
