Single-chip cameras add processing capability
Spurred by the market demand for a low-cost "camera on a chip," semiconductor and machine-vision companies are using CMOS technology to integrate more than a sensor on a piece of silicon. VLSI Vision (San Jose, CA) has already delivered the world`s first single-chip NTSC color camera, the VV6405. Designing a camera with the integrated circuit (IC) merely requires adding a single external crystal and single-rail 5-V power supply.
Like VLSI Vision, Motorola (Austin, TX) and Eastman Kodak (Rochester, NY) have recognized the mass market for CMOS imagers. At a September press briefing, held in Orlando, FL, the companies revealed that they are jointly developing CMOS-based imagers based on Motorola`s ImageMOS technology. According to preliminary data from Motorola, camera-on-a-chip products will include image sensors with on-chip analog-to-digital converters (ADCs), processors, and timing and clock generators.
Despite the industry enthusiasm, the first products are expected to be targeted at the mass markets of Internet teleconferencing and digital still cameras. At the same time, specialized smart sensors with on-chip signal processing are already appearing.
At September`s Vision Show, Integrated Vision Products (IVP; Linkoping, Sweden) described its MAPP 2200 IC that combines a 256 x 256-pixel image sensor, an ADC, and a 16-bit single-instruction multiple-data processor on a single chip. To permit rapid system deployment, IVP has integrated the 2200 sensor into a 2200 personal computer (PC). This system includes a camera, a PC interface, and development software.
According to Karl Gunnarson of Metolius (Redmond, WA), the US distributor for IVP products, typical performance benchmarks indicate that the camera can perform 3 ¥ 3 binary operators at 500 frames/s and median filters at 20 frames/s. At paper producer MoDo (Ornskoldsvik, Sweden), the 2200 PC system is being used in test equipment for measuring the gloss and gloss variations of paper.
In this measurement system, paper located on a test table is illuminated by a laser beam positioned at 45 to normal, and the light beam is reflected into the 2200 camera sensor. If the paper is completely flat, the angle of reflection is 45, and the beam is reflected into the middle of the sensor.
If there is a defect on the paper`s surface, the beam will be reflected at multiple angles, and a profile of the paper sample can be displayed. With a 0.1-mm-diameter laser beam, the system requires 10,000 measurements to cover a 1-cm square paper sample.
According to Gunnarson, the sensor`s image-processing rates are not the limiting factor in this application. "Because the test table moves 10 times a second, a 1-cm square paper sample takes 20 minutes to measure," he says. Electroencephalograms (EEGs) have long been used as a medical analysis tool for the diagnosis of brain disorders such as epilepsy. On EEG charts, epileptic seizures are characterized in the form of sharp waveform transients that are visually examined by medical experts.
To automate the charting process, researchers at the University Hospital of Valladolid (Real de Burgos, Spain) are proposing the use of wavelet transforms to improve the visual details of such EEG transients and, therefore, to make the epileptic conditions easier to diagnose accurately. "In the past, the short-time Fourier transform has been used to characterize the nonstationary EEG signals," states Robert Hornero of the university`s communications and engineering department. "But because segment length is directly related to frequency resolution, once a window has been chosen, the time-frequency resolution is fixed over the entire time-frequency plane," he says.
To overcome this problem, Hornero used several wavelet transforms in performing a multiresolution analysis of the signals. In the system developed by Hornero and his colleagues, a modified version of the Morlet waveform was applied to the data. The squared modulus of the wavelet transform or scalogram was then plotted to visualize the results.
"A version of the Morlet wavelet was chosen because you can recognize clearer differences between normal and epileptic EEGs in their scalograms. When plotting these scalograms, the benefit of using the wavelet technique is apparent. In patients with seizures, the EEG scalogram displays more regularity than in scalograms of patients without the disorder. Such data lots also convey information much more clearly than the EEG. For example, the periodicity of the EEG signal can be measured more easily, making the epileptic condition easier to detect. This process could save time, increase objectivity and uniformity, and enable quantification for research studies," says Hornero.
Better still, the results of Hornero`s work suggest that agents producing such seizures drive the brain into a stable periodic motion for certain cases of epilepsy. For more information, contact Roberto Hornero at [email protected].
