Always Optimizing
Improved image quality for video analytics gets help from wide dynamic range
- By John Monti
- Jan 03, 2008
Video analytics promises to revolutionize the use of security
cameras worldwide. For instance, intelligent video solutions
are being developed to identify a face from millions of
possibilities, pick out a suspect in a crowded stadium,
detect a group of potential terrorists or spot suspicious behavior on city
streets or in airports.
Equally intriguing are the non-security applications for video analytics,
which can improve business intelligence. For example, video analytics can
be used for inventory control, and both customer traffic pattern analysis
and demographic information are valuable for marketing intelligence.
effectiveness
depends on the quality of the video images. Data sufficient for
analysis requires that image sensors deliver wide dynamic range
(WDR), high signal-to-noise ratio (SNR), minimal image artifacts and
accurate color reproduction.
Expectations for video analytics have exceeded the capabilities of
charged-couple device (CCD) cameras. Generating consistently highquality
images—regardless of lighting conditions, temperature and other
environmental changes—while avoiding false alarms is difficult for traditional,
analog CCD-based security cameras. For these reasons, latest generation
analytics systems are relying on new digital image sensors.
Camera Limitations
Analog CCD cameras struggle to serve intelligent video applications
due to the limited dynamic range, which results in over-saturated
bright areas, under-exposed dark areas, poor color reproduction and
interlace artifacts. These all contribute to false alarms and missed
event detection. Solutions to overcome CCD deficiencies require filtering
and estimation that diminishes data accuracy, reduces resolution
and wastes valuable bandwidth with unusable data.
CCD sensors’ limited dynamic range leads to overflow of incident
light energy, resulting in image artifacts known as saturation, blooming
and vertical smearing. Saturation occurs when pixels in bright
areas are overexposed, causing them to turn completely white.
Blooming is the loss of color fidelity when the camera’s image sensor
captures a confined area with high illumination (e.g., the sun or a
bright object). Similarly, smearing occurs when large portions of
captured images are lost and appear as a vertical white band as
excess light energy overwhelms CCD read channels. Techniques to
overcome CCD image deficiencies result in loss of luminance, color
saturation and detail, delivering less accurate data for intelligent
video solutions.
Another problem with CCD technology is temporal video noise
caused by electron leakage on the sensor chip that appears as
crosstalk in the video stream. This random video noise diminishes
picture quality, contributes to false positives and increases the latency
of intelligent video algorithms, which perform real-time image
processing. The algorithms commonly mistake random sensor noise
for scene motion.
CCD-based cameras often experience interlace artifacts that
appear as saw-tooth edges around moving objects. This is due to the
traditional constraints of the NTSC and PAL television standards,
which have no use in video analytics applications.
In short, CCD limitations—image over saturation, video distortion
due to random noise and various pronounced image artifacts—
produce video streams that contain less data, forcing algorithms to
estimate or guess, causing false alarms or missed incidents. In addition,
extensive filtering techniques required to overcome CCD limitations
take up precious bandwidth, data storage and processing time,
all while reducing resolution.
Improving Video Analytics
A new category of security cameras offers WDR capabilities that
improve the accuracy of video images to deliver consistent, high-quality
data for video analytics. WDR refers to a camera’s ability to capture
images with the highest ratio between highlights and shadows. In other
words, WDR cameras can capture details and accurate color in both the
lightest and darkest portions of a scene simultaneously, even with highcontrast
lighting, strong backlight, glare, reflection and other uncontrolled
or variable lighting conditions.
WDR is measured in decibels. As a general rule, effective video analytics
requires a camera that provides 100 dB or greater dynamic range
to produce high-quality data with minimum noise (high SNR), accurate
color, excellent image quality and precise detail throughout all the lighting
ranges of a scene.
Video analytics algorithms must often estimate what the camera
“sees” in order to differentiate between foreground and background,
still and moving objects, or a target or event of interest from a video
artifact or other sensor errors.
WDR cameras deliver consistent, high-quality data for optimizing
video applications’ algorithms in order to produce accurate estimation
and identification by generating high-quality images with correct exposure
for the entire scene—both highlights and shadows—high-color
fidelity, sharp features with maximum detail, few or no image artifacts
and low video noise.
In addition to image quality and color rendition, all-digital WDR
cameras provide the best data for real-time network video analytics on
the edge—that is, analytics embedded in the camera. With all-digital
WDR cameras, there is no need to convert data from analog to digital,
and since video analytics is implemented at the edge, only significant
events are transmitted over the network, reducing traffic, false alarms,
time and data storage. Even if data is analyzed after it is transmitted to
a central server, WDR cameras deliver the best digital data ready for
analysis, reducing storage requirements.
New Possibilities
New digital sensors have improved the image quality of security camera
video to help meet the expectations of the security and analytics communities.
In particular, Pixim’s Digital Pixel System® (DPS) ultra-wide dynamic
range technology delivers image quality optimal for video analytics.
Cameras powered by Pixim’s specialized image processing chipsets
capture details and accurate color in both the lightest and darkest portions
of a scene simultaneously, even with high-contrast lighting, strong backlight,
glare, reflections and other uncontrolled or variable lighting conditions.
Cameras also minimize image artifacts common in CCD-based
cameras, such as vertical smearing, pixel blooming, under-exposure of
shadows, over saturation of highlights and interlace artifacts surrounding
moving objects.
By delivering the best WDR (maximum 120 dB), Pixim enables highquality
video images regardless of lighting or other environmental conditions.
This technology opens a range of possibilities for developers of
video analytics applications, makers of cameras and the customers who
use them.
DPS technology’s ultra-wide dynamic range, high resolution, accurate
color and minimal image artifacts can directly improve security in a variety
of important applications, such as financial services institutions, retail
stores, casinos, perimeters and borders, schools and campuses, correctional
facilities and a variety of transportation-related environments, including
airports, seaports, trains, light rails and buses. Beyond these security applications,
high-quality images also can enable video analytics and improve
business intelligence.