Most living creatures, including humans, have two eyes. This is probably the result of evolutionary processes over many eons that determined two eyes to be most effective for survival. This fact led OMRON to conclude that vision sensors should also have two "eyes," or two cameras.

To gather information from a person or object, it is necessary to assess it three-dimensionally. However, it would take an incredibly large, intricate system to process the huge volume of information that a three-dimensional object contains. Fortunately, every object has a specific viewing direction that makes it easiest to identify. For example, a car can be identified when viewed from the side much more easily than when viewed from the front or above.

Shapes of a car viewed from different directions

The same applies to machines as well; machines also find it easiest to recognize a car from its side-view shape. Still, it may not always be possible to install a camera in the direction that makes it easiest to recognize the three-dimensional shape of all objects. In addressing this issue, OMRON developed the innovative vision sensing technology dubbed "Silhouette Vision." The system uses two cameras to capture a stereo image and extracts the characteristic form (silhouette) of the object from that image in a highly efficient, resource-conserving manner.

Imagine that a car passes in front of you. As human beings, we can recognize its approximate shape and size just by looking at it. If you were asked to draw an outline of the car from the front, top and side view, you could probably do it just from your memory of the car and your ability to estimate what its shape might look like from various angles.

Silhouette Vision is designed to let machines utilize the same processes that people use instinctively. By deploying two cameras, a machine can detect the height of the object and how far it is from the viewing point, and recognize it three-dimensionally. By so doing, the machine can accurately determine and project the silhouette form of the object, which is viewed from the most appropriate direction (for example, the side view in the case of a car). In other words, Silhouette Vision technology can capture the essential three-dimensional form of an object from the two-dimensional camera image and ultimately convert it into a two-dimensional silhouette image for processing. Thus, the huge amounts of information that make up a three-dimensional image can be captured in a simplified and efficient manner, which enables a compact, resource-saving hardware design.

A stereo camera system (using two cameras) can capture an object's height and depth.

Keeping Two Eyes Out for Traffic Conditions

Return to Vision Sensing Top