Improved predictive analytics correctly identify safety hazards
Buses present high security risks because their public nature makes
it difficult to screen each and every passenger. Speaking to the House
Homeland Security subcommittee on Transportation Security, Greyhound Bus
Lines Chief Executive Officer William Blankenship told the panel that
an estimated 500,000 items brought on Greyhound buses each year could in
effect be used to overtake drivers.
Coupled with this is the fact that less than 2 percent of the
Transportation Security Administration budget goes toward protecting
surface transportation. This puts the burden of safety squarely on local
law enforcement and individual transit agencies, which are facing
budget cuts of their own.
All of this means that bus operators are increasing their use of
technology to enhance security in cost-effective and unobtrusive ways.
One of the biggest technology trends we are seeing in bus transportation
security is the use of predictive analytics — identifying possible
safety hazards and correcting them before an incident occurs.
At their core, predictive analytics are sophisticated software
algorithms that create alarms based on “if-then” scenarios. For example,
if a security camera captures video of someone dropping a large package
into a trash can at a bus station, the system then sends an alert to
authorities who can investigate the activity. Or, if a piece of baggage
is loaded onto a bus but its owner does not board, it then notifies the
police of a suspicious package.
Of course, the algorithms are typically much more complex than this
and can operate based on any number of triggers. The complexity of the
technology means that all of the data under evaluation by the software
needs to be as accurate as possible in order to avoid generating false
alarms, or “false positives.” This has been a common problem with
analytics to date and is an issue the security industry is constantly
striving to improve.
One way that predictive analytics is improving is through better
image quality in security cameras. The vast majority of analytics rely
on visual data from IP-based security cameras, like their capability to
detect motion or identify abandoned packages. The images need to be
clear so the software can accurately analyze what is happening in each
frame. Imagine the waste of time and money if security guards are
dispatched to a bus depot to check a motion detection alert after hours,
only to find that an opossum or a tree branch has tripped the alarm.
Unfortunately, this has become a common reality with poor quality
images fed into analytics software from security cameras. We have all
seen security footage that is too grainy, too dark or washed out to be
of any use.
Lighting is the most common cause of poor video quality. Lighting
conditions change dramatically throughout the day and even season by
season. A camera facing a bus station might have ideal lighting
conditions for only a few hours each day. For the rest of the time there
are likely to be issues with back lighting, darkness or glare that
seriously erode picture quality. In these cases, it can be quite
difficult for the analytics software to distinguish between a person and
a pesky opossum.
Thankfully, IP camera technologies like those under development by
Sony are now addressing these concerns by improving the “wide dynamic
range” of IP cameras. Video taken at various shutter speeds combine into
a single video stream that makes images more visible even under
extremely high-contrast lighting conditions. This improves the quality
of the visual data that is fed into the analytics software and reduces
the number of false alarms.
Wide dynamic range takes image quality beyond the mega-pixel trend in
security cameras. Today, there is a common belief in security cameras
that with more megapixels comes better image quality. While that is
partly true, increasing the number of megapixels on a camera does
nothing to improve how well that camera will perform under a wide
variety of lighting conditions. A 10-megapixel picture of a figure that
is unidentifiable due to extreme backlighting will do nothing to improve
the accuracy of analytics software. On the other hand, a five-megapixel
camera with excellent dynamic range can provide much more valuable
In addition to improving the dynamic range of IP cameras, reducing
the “noise” in low-light or fast-moving video enhances image quality and
enables analytics software to be more accurate. In video transmission,
noise refers to the level of static or blur in an image. Static often
results from images that are poorly lit, while blur is a common problem
with fast-moving objects against a still background. When using
analytics, this noise often results in false alarms because software
algorithms can interpret the pixel changes as motion.
Digital noise reduction technologies such as Sony’s XDNR feature
apply filters to video that decrease static or blur. This helps IP
security cameras and analytics software distinguish true motion from
Unfortunately, there are still no standards in the security industry
that dictate minimum performance for dynamic range and noise reduction
in IP cameras. Prospective users need to conduct thorough research to
determine which technology will produce the best image quality for their
specific environment and lighting conditions. However, as the
technology continues to improve and standards evolve, we expect that the
use of predictive analytics in bus transportation systems will only
Links that further explain the importance of wide dynamic range in video surveillance:
Explanation of dynamic range and wide dynamic range technologies
Actual video of backlit figure walking on train platform:
Mark Collett is general manager of Sony Electronics Security
Systems Division. He has more than 20 years of security and technology
industry experience. Collett is on the Board of Directors for the
Security Industry Association (SIA) and he is a frequent speaker at
security industry events.