Keeping an eye on security
Wide dynamic range image sensor-processor improves security video and other applications
January 16, 2007
From: Homeland Security Asia
By: John Monti
The extended dynamic range, fast readout speed, system-on-chip
integration and low power operation of Digital Pixel System (DPS)
technology from Pixim represents a substantial advance over existing
image capture and processing technologies. Enhancing video image
capture for many applications, DPS is particularly useful for security
cameras that must provide detail over a wide range of illumination
conditions.
Solid-state image sensor technology dates back to the invention of the
first charge coupled device (CCD) in the late 1960s. The 1990s saw the
introduction of the CMOS active pixel sensor (APS), followed by the
development of the DPS platform. The invention of DPS technology grew
out of more than eight years of research at Stanford University. Pixim
subsequently commercialized DPS three years later. Pixim's recently
announced 'Orca' D2500 chipset provides all of the functions necessary
to build advanced surveillance cameras and emerging imaging
applications in two small Ball Grid Array devices.
The Digital Pixel Systems architecture and tightly-coupled imaging
software provide superior image quality even under widely varying light
conditions and wide dynamic range scenes that have both dark and bright
areas. DPS provides dramatically better wide dynamic range images than
charge-coupled device (CCDs) or CMOS active pixel sensors (APS) used in
similar video camera applications.
The highly integrated two-chip set is comprised of a digital image
sensor chip and a digital image processor chip. Layout of chip
interconnections and user interface is straightforward. The user
interface includes a customizable menu-driven on-screen display (OSD),
switches and potentiometers, in any combination. Manufacturers define
which of many settings are available for user control and which are
fixed at the factory. Fixed settings can be locked so that they may not
be identified or changed in the field, if desired. Additionally, DPS
technology offers:
- Programmable exposure controls, configurable noise reduction, and greater dynamic range.
- Digital video output and/or analog composite video output.
- Reduced fixed pattern noise problems commonly associated with other sensors.
- Digital pan / tilt / zoom, automatic white balance, and color
correction, among other capabilities, using digital signal processing
provided in the image processor chip.
DPS Image capture and processing
DPS technology converts the quantity of light striking each picture
element (pixel) to a digital value at the earliest possible point: at
the pixel itself. An analog-to-digital converter (ADC) is designed into
each pixel, and is operated simultaneously with all other ADCs in every
pixel of the sensor. This pixel-level ADC architecture permits the use
of many highly parallel low-speed circuits, operating close to where
the photodiode signals are generated. This is key to optimizing the
signal-to-noise ratio (SNR) for each pixel.
The DPS system uses the individual ADCs in each pixel to perform
non-destructive correlated double sampling (CDS) at each pixel. DPS
uses this capability to sample the growing light intensity at each
pixel many times during each image capture period. This allows exposure
level of each pixel to be determined by the rate of change of charge
collected rather than only its absolute magnitude. Each pixel is also
provided with an adjustable offset cancellation gain amplifier to
assure uniform response throughout the sensor array. These innovations
greatly reduce noticeable fixed pattern noise problems commonly
associated with the column-level ADC used on APS sensors.
Because DPS sensors are digital, pixel readout is much faster and more
accurate. Each sample of the digital image is captured in on-chip RAM.
The high bandwidth provided by tightly coupled local memory is used to
achieve its superior high dynamic range. This approach is not practical
for CCD or APS sensors because of their reliance on analog readout
circuitry. This is not a problem with DPS, which greatly benefits from
the digital sampling performed on each pixel.
Wide dynamic range
Wide dynamic range is essential for capturing image detail at all light
levels. Today's surveillance cameras are plagued by dynamic range
problems in a typical 24 hour day, due to severe reflections, glare,
car headlights, and direct sunlight. The wide dynamic range achieved by
DPS is realized by a patented non-destructive multi-sampling image
capture capability, and advanced image-processing algorithms.
Dynamic range is the ratio of the brightest image that can be captured
by the imaging system to the darkest image that can be captured. Light
intensity greater than the brightest possible image will cause the
sensor to saturate, while light intensity less than the darkest
possible image will not register on the sensor. Both of these
conditions distort the image, hiding potentially vital information that
lies outside the dynamic range of the sensor.
When an exposure begins, each pixel is charged at a rate that is
proportional to the intensity of the light that strikes it. A stronger
light source will charge a pixel more quickly than a weaker light
source. Existing analog technology typically uses a single exposure
time for all pixels. At the end of the exposure, the camera will sense
the total charge accumulated in each pixel. But that means some pixels
(the brighter ones) may be overexposed while others (the darker ones)
may be underexposed.
DPS overcomes this limitation as follows: with DPS, the light striking
each pixel is sampled multiple times during the exposure period. DPS
analyzes how quickly each pixel is being charged by the light striking
it. This way, DPS measures light intensity by a combination of the rate
at which the charge grows as well as the total charge accumulated
during an entire exposure.
Specifically, the DPS system records the length of time required to
nearly saturate each pixel. Pixels exposed to bright illumination will
tend to saturate more quickly than other pixels. DPS determines for
each pixel whether it will saturate before the next sample. If a pixel
would saturate, then its elapsed exposure time is stored in memory,
together with its current intensity of charge.
The advantage of this approach can be appreciated when one realizes
that the entire range of each individual pixel, as well as the rate of
change of the pixel charge, is used to form the resulting image,
significantly increasing the dynamic range and SNR that is captured.
Other technologies only measure the pixel value, not its rate of change.
DPS also provides improved color performance not available with other
sensor technologies: the data recorded by each pixel is of very high
quality, both in terms of accuracy and precision. High data quality
allows the DPS image processing algorithms to render excellent fidelity
for all colors and intensities. In surveillance applications, color
accuracy is critical to forensic analysis of stored video once an
incident has been recorded.
DPS provides a fast global electronic shutter to capture bright lights
and produces images that do not exhibit rolling shutter artifacts
common in APS sensors, or interlace artifacts common to CCD sensors.
Since multi-sampling is fundamental to DPS and is included in the basic
firmware, no programming is required by developers to achieve this
level of quality.
Improving signal to noise ratio
The signal-to-noise ratio (SNR) is a measure of a captured image's
immunity to noise interference. Peak SNR is the ratio of the strongest
recordable intensity without saturation to the background noise. A
camera with higher SNR typically produces better (less noisy) video in
darker scenes. Photodiodes generally have better SNR when they are
charged to more than half of their total capacity. But they cannot be
charged beyond saturation. In other (analog) sensor technologies, high
contrast portions of an image cause saturation of the photodiode and
"blooming" in adjacent pixels.
The SNR in a DPS sensor is greater because each pixel is measured at
its maximum value just prior to saturation. For example, for the
photograph shown in Figure 2, the best exposure times are: T2 for the
apple, T4 for the region below the cup, and T6 for the shadowed portion
of the car because it is at these times that the photocurrent produced
by these pixels is highest prior to saturation or the end of the
exposure interval. Therefore, these are the times when the image is the
most accurately represented by these pixels.
DPS reduces noise in the sensor in a number of ways. First, a negative
feedback unity gain amplifier in each pixel eliminates any offset
voltage, resulting in much greater uniformity throughout the sensor
array. To minimize reset noise, each pixel value is read
non-destructively at the beginning of the exposure, and this value is
later subtracted from the final measured value for the pixel. This
non-destructive method of correlated double sampling is unique -- most
other sensors must read the CDS value, reset the sensor, then capture a
new value. But the random background noise change from the initial
captured CDS value introduces distortion that DPS avoids. This reduces
noise that would be detectable to the human eye.
The high quality of Pixim's pixels provides high signal levels and low
noise levels, resulting in excellent inherent SNR characteristics. This
can provide sensitivity on par with CCDs, and an order of magnitude
sensitivity gain over APS sensors, while providing extended dynamic
range and excellent image quality.
Pixim DPS image sensing technology
Because DPS technology integrates the ADC into the pixel, it can be
manufactured using leading edge semiconductor processes. The pixel
array has significantly higher noise immunity than analog sensors
because DPS technology employs a digital readout from each pixel.
Additional image processing and camera functions implemented by Pixim
provide a complete imaging solution in high-volume, commercially
available chipsets, the latest of which is the Orca product family.
DPS image systems integrate sensing, memory and processing functions
into two chips. This is especially important for imaging systems that
require significant processing, where quality of output is crucial, and
where small size, low power and time-to-market are important. Until
now, analog-to-digital conversion could only be integrated at the chip
or column level. Both approaches are common in analog sensor solutions.
For the chip-level approach, a single conventional high-speed ADC is
integrated with the sensor. For the column-level approach, one or more
columns of the pixel array has a dedicated ADC. The ADCs are operated
in parallel and, therefore, low-to-medium speed conversion techniques
must be used (e.g., single-slope, algorithmic, successive
approximation, or over-sampling).
By having separate ADCs for each pixel operating in parallel, the ADCs
can operate at very low speed, operating at a few thousand samples per
second. This lessens noise and reduces power requirements of DPS
sensors.
The large bit stream used to read the ADCs is supported by on-chip RAM.
These features enable much faster and more accurate readout
characteristics. The DPS sensor consists of a digital pixel sensor
array, ADCs and RAM. A separate image processor chip incorporates
digital signal processing (DSP), a second video frame buffer, and I/O.
The sensor core is powered by a single low-voltage power supply.
Manufacturing is simpler than for CCDs and greater miniaturization is
realized. These advantages will continue to increase in the future in
accordance with Moore's Law.
Pixel-level ADC: The DPS pixel-level ADC architecture permits
the use of low-speed conversion; the ADCs operate close to where the
photodiode signals are generated. This optimizes signal-to-noise ratio
(SNR) and power consumption beyond any sensors currently available. The
large number of independent, small ADCs significantly reduces
noticeable fixed pattern noise problems commonly associated with the
column-level ADC variation on APS sensors. Finally, because pixel
signals are available to ADCs at all times, the number and timing of
images—as well as the number of bits from each image—can be freely
chosen. This offers important advantages, such as the ability to
optimize the image capture and processing to the scene characteristics.
All of these features have been implemented in Pixim DPS systems.
DPS pixel readout: Each pixel's ADC feeds into a RAM-speed
bit-serial readout. The multi-channel bit-serial design reduces the ADC
output data rate, allowing it to use standard Nyquist-rate conversion
instead of over-sampling. It facilitates DPS dynamic range enhancement
via multiple sampling, and reduces non-uniformity by globally
distributing the ADC control and clock signals, and by performing
auto-zeroing. The implementation consists of a two-dimensional array of
pixel blocks, a row decoder and column sense amplifiers and latches.
Each pixel block comprises one or more photo detectors sharing an ADC
channel. The captured analog pixel values are digitized in parallel,
one bit at a time. Each latched set of bits, using the row decoder and
column sense amplifiers, forms a bit-plane that is read out in a manner
similar to a standard digital memory. As depicted in Figure 6, a set of
bit planes constituting an m-bit frame of the data is collected during
a single exposure window. In this example, the digitized value of the
uppermost right-hand corner pixel is 1001...1. Note that this image
output format is quite different from the raster-scan format commonly
used in CCD and APS devices.
The DPS system: Beyond the fundamental DPS technology, Pixim has
integrated functionality into the silicon and software to produce a
comprehensive system solution. Developers can build high-performance
video surveillance cameras with straightforward hardware design and
simple configuration using the Pixim Configuration Language (PCL) to
control the hardware interfaces, develop menus, and control the imaging
behavior of the camera. No signal processing software development is
necessary. Pixim also provides complete reference designs and a variety
of camera configurations for very quick time to market.
John Monti, the founding executive of Pixim, Inc., is Vice President of
Sales and Marketing. He draws on experience in the fields of
semiconductor applications, marketing and system-on-chip solutions, and
is a published writer and speaker. For more information, e-mail:
monti@pixim.com.