# Binary Vision Sensors

Wednesday, May 2nd, 2018 -

## Binary Vision Sensors

This class of sensor produces images whose pixel (pixel = picture element) values are either a black or white luminosity level equivalent to a logic 0 or 1, hence the name ‘binary’. Figure 1 shows a typical block diagram of a binary vision sensor:

Figure 1. Block diagram of typical binary vision sensor

The complete picture is therefore only a series of logic 1 and O. This allows easy distinction of dark objects on light background (and vice versa) and, in view of the low visual data content, fast image data manipulation such as object perimeter and/or area calculations.

In the example shown the vision system has learnt the characteristics of three objects, based on the number of holes, object size and shape features. Thereafter, whenever the vision system is presented with an image containĀ­ ing one of these objects it will be able to ‘recognize’ it, that is acknowledge its presence within the field of view. This is achieved by comparing the features of the object in the new image with those stored in memory, looking for a match. In practice a certain amount of tolerance needs to be built into the matching algorithm because the sensing process is not perfect and will not always produce the same output for the same object, so that a ‘perfect’ match would hardly ever be possible. Too much tolerance, however, may result in the vision system ‘recognizing’ two similar objects as being one and the same.

Figure 2 shows the typical output of a complete binary vision system:

Figure 2. Typical example of an object recognition program based on a binary image

This technique of image processing is sometimes referred to as the SRI method (or one of its derivatives) because it was first introduced by Rosen et al. at the Stanford Research Institute (now SRI International) and imposes severe restrictions on the lighting conditions and on the position of the object for a successful operation. The object must, in fact, show up as an isolated silhouette and its stable states must be known in advance (that is the object can be rotated within the image plane but not tilted with respect to it). Thus overlapping objects and situations with any 3-D freedom of movement are difficult to deal with using binary vision sensors.