In the USA, as in Europe, there is still considerable interest in 2-D robot vision, both in terms of an effective and practical, though limited, implementation of visual sensory feedback-there is in fact a school of thought that prefers 2-D to 3-D vision systems in view of their simpler and more trouble-free operation (Keller, 1983; Corby, 1983)-and as the input stage for ‘model-based’ vision processing systems that aim to derive a 3-D image from the 2-D data using a suitable model of the object in question (Meyer, 1984). The research emphasis, however, is on 3-D robot vision as a means of achieving faster, more accurate and more flexible robot operation.
2-D vision sensors are based on optical array transducers (both vacuum and solid-state types, such as camera tubes, CCD, DRAM, and photodiode arrays), and specially designed artificial lighting. Some ‘intelligent’ 2-D vision sensors also possess limited computer power to enable them to carry out a certain amount of picture pre-processing. This may be used both to relieve the data-processing burden on the main control computer and to interact with the illumination system (e.g. to control the lighting level) and so provide the best possible picture quality to the main computer.
Most optical array transducers (e.g. vacuum and solid state cameras) have the capability, though via different means, to provide a ‘grey level’ output image. The only exception to this guideline is the DRAM camera, being based on a memory device, produces an inherently binary output image. It should be pointed out, however, that a grey level image can be produced with a binary device at the expense of the image acquisition time by overlaying several binary images obtained with different exposure times.
The lighting conditions and/or the computing power limitations may, however, be such that the transducer output needs to be interpreted as binary anyway. This is the case for through illumination which, for opaque objects, helps to produce images with only two distinct grey levels. Another such example is the use of microcomputers for robot control and image processing: these devices cannot, in fact, easily handle grey level images because of their large memory requirement and therefore often need to threshold the transducer output (a 256 x 256 image with a 8-bit grey level resolution would in fact require 65,536 bytes of memory to store it). 2-D vision sensors can be therefore divided into two main groups:
- binary vision sensors
- grey level vision sensors