Page 482 - The Mechatronics Handbook
P. 482

0066_frame_C19  Page 104  Wednesday, January 9, 2002  5:32 PM













































                       FIGURE 19.83  HERMIES IIB employed an active stereoscopic ranging system with an external laser source that
                       could be used to designate objects of interest in the video image (courtesy Oak Ridge National Laboratory).

                       Active Stereoscopic
                         Due to the computationally intensive complexities and associated resources required for establishing
                       correspondence, passive stereoscopic methods were initially limited in practical embodiments to very
                       simple scenes (Blais et al., 1988). One way around these problems is to employ an active source in
                       conjunction with a pair of stereo cameras. This active illumination greatly improves system performance
                       when viewing scenes with limited contrast. Identification of the light spot becomes a trivial matter; a
                       video frame representing a scene illuminated by the source is subtracted from a subsequent frame of the
                       same image with the light source deactivated. Simple thresholding of the resultant difference image
                       quickly isolates the region of active illumination. This process is performed in rapid sequence for both
                       cameras, and the lateral displacement of the centroid of the spot is then determined.
                         Alignment between the source and cameras is not critical in active stereoscopic ranging systems; in
                       fact, the source does not even have to be located on board the robot. For example, Kilough and Hamel
                       (1989) describe two innovative configurations using external sources for use with the robot HERMIES
                       IIB, built at Oak Ridge National Laboratory. A pair of wide-angle black-and-white CCD cameras are
                       mounted on a pan-and-tilt mechanism atop the robot’s head, as shown in  Fig. 19.83.  Analog video
                       outputs from the cameras are digitized by a frame grabber into a pair of 512 by 384-pixel arrays, with
                       offboard image processing performed by a Hypercube at a scaled-down resolution of 256 by 256. The
                       initial application of the vision system was to provide control of a pair of robotic arms (from the Heathkit
                       HERO-1 robot) employed on HERMIES.
                         To accomplish this task, a near-infrared LED is attached to the end of the HERO-1 arm near the
                       manipulator and oriented so as to be visible within the  field of view of the stereo camera pair. A
                       sequence of images is then taken by each camera, with the LED first on and then off. The off represen-
                       tations are subtracted from the on representations, leaving a pair of difference images, each comprised
                       of a single bright dot representing the location of the LED. The centroids of the dots are calculated to


                       ©2002 CRC Press LLC
   477   478   479   480   481   482   483   484   485   486   487