Page 430 - Sensors and Control Systems in Manufacturing
P. 430

384
                                    Ei g h t
                           Cha p te r

                          the image such as object edges, alignment, regions, boundaries, colors,
                          and the absence or presence of vital components.
                             Once the image is analyzed, the vision sensing system can inter-
                          pret what the image represents so that the robot can continue its task.
                          In robot vision execution, design considerations entail cost, speed,
                          accuracy, and reliability.
                          8.3.2  Robot Vision and Human Vision
                          Given that robot vision systems typically execute only part of what is
                          normally called seeing, some similarities with human vision never-
                          theless arise. Other similarities arise in the “hardware” of human and
                          robot visual systems.
                             The similarities in hardware could include an analogy between
                          eyes and video cameras—both have lenses to focus an image on a
                          sensitive “retina” that produces a visual signal interpreted elsewhere.
                          In both human and robot vision this signal is passed to a device that
                          can remember important aspects of the image for a time, perform
                          specialized image processing functions to extract important informa-
                          tion from the raw visual signal, and analyze the image in a more gen-
                          eral way.
                             Some similarities in performance follow. Human and robot vision
                          work well only where lighting is good. Both can be confused by shad-
                          ows, glare, and cryptic color patterns. Both combine size and distance
                          judgments, tending to underestimate size when they underestimate
                          distance, for instance.
                             However, humans and machines have far more differences than
                          similarities. A human retina contains several million receptors, con-
                          stantly sending visual signals to the brain. Even the more limited
                          video camera gathers over 7 Mb of visual information per second.
                          Many of the surprising aspects of machine vision arise from the need
                          to reduce this massive flow of data so it can be analyzed by a com-
                          puter system.
                             Machine-vision systems normally only detect, identify, and
                          locate objects, ignoring many of the other visual functions. However,
                          they perform this restricted set of functions very well, locating and
                          even measuring objects in a field of view more accurately than any
                          human can.

                          8.3.3  Robot Vision and Visual Tasks
                          Several standard visual tasks are performed by robot-vision systems.
                          These tasks include recognizing when certain objects are in the field
                          of view, determining the location of visible objects, assisting a robot
                          hand with pickup and placement, and inspecting known objects for
                          the presence of certain characteristics (usually specific flaws in manu-
                          facture) (Fig. 8.2a).  Figure 8.2b illustrates advanced vision inspection
                          and measurement system capturing a protein like string consisted of
   425   426   427   428   429   430   431   432   433   434   435