Skip to Main Content
Robotic vision systems make use of image shapes to identify and locate objects in the workspace of the robot. For objects observed by a normal lens system, and centered in the image, shape parameters can be reliably extracted and compared to reference models by a number of well-known techniques. In many applications, it appears desirable to mount the camera very close to the workspace, and to use a wide angle lens. This overcomes problems caused by congested manufacturing workstations, and allows a wide field of vision. Unfortunately, the shapes of objects in the peripheral area of camera images have a distortion, inherent to planar projective mapping, which increases with increasing angles measured from the optical axis. A significant portion of the image area from a wide angle lens falls in this category. When the objects contain depth, this distortion alters angles and sizes of solid objects. This paper presents a transformation that corrects the image shape of any selected object or region in a wide angle digital image. It performs the transformation by defining a virtual camera to "take a picture" of the image with the desired orientation. A new, simulated, image is formed with correct shapes and angles making shape recognition easier. Examples of the implementation of this algorithm are presented.
Robotics and Automation. Proceedings. 1984 IEEE International Conference on (Volume:1 )
Date of Conference: Mar 1984