How do you think the new GigE standards will influence the machine vision industry?
Respond or ask your question now!
Q&A With Len Yencharis
An interview with Ricardo Motta, Chief Technology Officer & Vice President of Imaging Systems at Pixim, Inc., regarding a new technology that can digitize photons right at the sensor...
Len Yencharis, Advanced Imaging: I would like you to compare briefly where this development stands with respect to other efforts. If Pixim were able to perform automatic image recognition, could that technology be used for automated image analysis in such applications as medical, scientific R&D and real-time image processing?
Ricardo Motta, Pixim: Pixim has developed a high-quality video chip set called the D2000 Video Imaging System, based on imaging technology developed at Stanford. It captures, processes and produces video signals in a variety of interfaces, including SDI suitable for image recognition. The D2000 does not have the ability to perform automatic image recognition in the chip set, although there are simple ways to interface to a microprocessor to perform the recognition on high-quality image data. The D2000 has an internal ARM processor used primarily for camera control, but can also do some image analysis such as activity detection. The D2000 brings a lot of benefit to these image analysis applications as explained below. Pixim's DPS (Digital Pixel System™) technology brings advantages to pattern recognition and image analysis in general. A digital image sensor first digitizes the image data at the pixel, eliminating many of the issues related to read-out speed and scalability of the sensor. One of the unique benefits of DPS is the elimination of pixel jitter. The images from our first video product look very stable. In fact, we sometimes have to nudge the camera or move something in front of it to make sure it is live since there is absolutely no jitter in the image. Other key benefits that DPS brings to pattern recognition are extended dynamic range, progressive readout, square pixel and high frame rate. Pixim's first video chipset, the D2000 Video Imaging System, combines these advantages with many easy to use features such as graphic overlays and the ability to switch between PAL or NTSC (the product synthesizes the fields in hardware from the progressive 60 FPS data, and can produce PAL or NTSC outputs using the same sensor and logic) and a variety of outputs such as SDI, component, YC and CVBS. These are significant advantages for applications in the security, vision, medical and R&D fields, which will only get better with time.
Yencharis: How do you test for such performance issues as sampling (oversampling capability), parallel a/d data conversion rate and column read noise (differentially, or is FPN not a factor?)
Motta: FPN is, by far, the biggest issue faced by current CMOS designs. Pixim's per-pixel conversion eliminates familiar FPN patterns such as column or row artifacts. We have addressed the remaining FPN by a combination of sampling, CDS and image processing. The architecture's low noise readout, combined with non-destructive reset and high readout speed, allows our digital image sensor chip to measure the reset values of all pixels in every capture, resulting in a very effective CDS. Since Pixim started testing and qualifying the D2000 Video Imaging System, we have developed a number of new techniques and metrics. For instance, in the DPS platform, much of the circuitry can be tested without light, since our chip has an embedded DAC and we can change the reset voltage and measure the reset value for every pixel. Any shorts or defects in the circuit can be identified in a logical test. The quality criteria for the optical testing are not very different from other sensors, with attention paid to the signal sources, like the QE, and the noise sources, like read noise.