In my research laboratory, we are working on building new camera-like sensing systems that specialize in detecting hard-to-find objects or patterns and which can also be easily reconfigured to detect a variety of different targets.
My research involves using new technologies to develop sensing systems that are fast, adaptive, and compact. We are working on fabrication technologies that enable us to make very small movable mirrors and customized light detectors (for measuring the location, color, and brightness of the light) that are actual physical components of the sensing system. We have devised a scheme for filtering in which we convert the light information into an electrical signal that is easy to filter using a separate set of electrical signals that we create and control. Clever filtering of an image can help us notice colors we are looking for or shapes that are tough to discern. It is easy to reprogram these electrical signals (changing the filters), so the same "smart camera" can quickly search for whatever targets we want.
This concept, a camera-like system that can preferentially sense predefined targets with the ability to quickly change sensitivity, is referred to as "adaptive spectral sensing." "Adaptive" indicates an ability to program the system to detect different things by changing the way it filters the input. The word "spectral" describes what is being measured: essentially, different colors of light. "Sensing" is the process of receiving, measuring, and analyzing some sort of physical input, or in this case, light.
A video camera can be likened to an artificial vision system. Like the human eye, a camera captures an image of what it sees; unlike the human vision system, however, most cameras cannot pick out particular things that you want to find. For example, suppose you are looking for someone you know in a large crowd, and you know that that person is wearing a red sweatshirt. You might have noticed that, as you search through the sea of colors, your brain pays attention to people wearing red and ignores others. This is an example of filtering; your brain is sorting out a very large amount of visual data (everything you can see in the crowd), ignoring useless information (somebody in a black shirt, perhaps), and paying attention to potentially useful information (a person wearing red). The more you know about your target, the better your filter is, and the easier it is to find what you want.
Now consider a video camera that is mounted for security in a convenience store. The camera records a picture representing what it sees; this is called imaging. It gathers visual information just like your eyes do. However, current video systems are not optimized to identify objects, such as human faces. Cameras record information from all colors across the entire scene with the same level of detail, regardless of content. For example, a robber's face and a cereal box would be treated with equal consideration by the camera, even though the face is a more important target. This is unfortunate but not surprising given that the camera has not been designed to image faces better than other objects.
But cameras can be designed, it turns out, to image certain things better than others. Filtering the input is a critical step. We want to keep useful information about colors or shapes (objects that are red colored or shaped like a human face) and throw away other information (details of a cereal box). How can this be done? In any sensing system, light is detected, converted into an electronic signal of some sort, and then analyzed. A filter that has the properties we want can be applied before detection, when the information is in the form of light, or after it has been detected, when it is in the form of electrical current. Either way, the filter adjusts the system such that important information is preserved while extraneous information is not. In the convenience store, filters matching human face characteristics could allow the camera system to record less information about unimportant objects (cereal boxes) and to record more about human faces. For example, the background could be imaged in black and white only, whereas a human face could be imaged in full color.
When we know ahead of time exactly what we want to image, we can simply design and build the appropriate device. In many applications, however, it might not only be difficult to detect what we are looking for, but the object might also change over time. We may want to look for one thing now and another thing later on, and it is not practical to build a separate camera for each type of object that we might want to image. For example, there are many satellites orbiting and imaging Earth looking for objects of interest. Some targets, like military equipment, are often hard to find because they can be concealed. Sensing systems on the satellites, much like cameras, can be designed to image certain types of things but the desired targets might change as the satellite operators wish to detect different objects. Operators may be interested in initially detecting a military vehicle, then a rocket launch, and finally a hazardous cloud of chemicals. Being able to change the configuration of the satellite's sensing systems at will would significantly improve the detection of those different objects. One sensing system could be used to accomplish a wide variety of tasks efficiently and economically because of its adaptability.
If we can continue to improve the sensing systems we are working on,
it might be possible for "smart cameras" to be used in a variety
of everyday applications, such as testing for certain elements of blood
(e.g. sugar levels), checking for the purity of water supplies (e.g. minerals
or bacteria), or factory workers keeping a lookout for dangerous chemicals
in the air. This kind of research requires imagination and creativity
as well as knowledge of scientific principles. We are encouraged to invent
new ways to solve these problems, and that makes this project exciting.
|Modified 15 January 2003 * Contact Us|