Over the past decade and especially after incidents like
“9/11” and "26/11", the emphasis on video based surveillance technologies has
really picked up. The use of IP surveillance cameras and CCTVs have increased
across both government and private enterprises. However, deploying these devices
alone isn't enough. The next challenge lies in the physical monitoring of videos
being grabbed by them. There's bound to be human error while monitoring
surveillance cameras 24x7. That's where video analytics technology comes to the
rescue. Let's understand this with the help of two scenarios--first without and
then with the use of video analytics:
Scenario 1: A terrorist scales the boundary wall of
a petroleum company's oil depot and plants bombs at the oil storage tanks, which
cause vast damage. The installed CCTV or IP surviellance cameras grab the
footage of this entire activity, which are later analyzed by the security
agencies to investigate further.
Scenario 2: A terrorist scales the boundary wall as
in the first case. This time, there are motion-detection surveillance cameras
installed, which detect the suspicious movement, and trigger an alarm to notify
the security personnel about the location of intrusion. The security people
reach the location on time to prevent the mishap from happening.
The above example very clearly illustrates the power of
video analytics solutions, clearly indicating that simple surveillance
techniques are just not enough today. You need video analytics solutions to make
sense of the video being grabbed and take necessary action 'in time'. They can
be used to provide real-time security, with an emphasis on preventing security
incidents instead of recording them for post analysis. Let's understand how it
works.
Video Analytics and its usage
Video Analytics is a way to do automated analysis of video content for
specific data, behavior, objects or any other user-defined events. The
technology is primarily used for physical security and business intelligence
apps. The video analytics software capability ranges from motion detection to
complex algorithms for detection of people, objects and their behavior. Similar
to human vision, which is perceptual and cognitive, video analytics also employ
complex vision algorithms to enable the solution to see and interpret from the
video content. The objective of video analytics is to understand the scene and
not just do motion detection. In case of motion detection cameras, even a single
pixel change can trigger an alarm, while systems with video analytics can ignore
motions from tree foliage, rain or snow, and can concentrate on scene change due
to human or any other object's motion.
Video Analytics softwares like CyeWeb from NovoSun that takes the video feeds from IP surveillance cameras and analyzes them for user-defined incidents |
Video analytics can be deployed to automate security
surveillance, be it through CCTV cameras or IP surveillance cameras. For
instance, video feeds from cameras can be continuously analyzed to detect
presence of people and interpret their activities. So, suspicious activity or
people moving in unauthorized areas would automatically raise a flag and notify
the security personnel. In case of business intelligence apps, video analytics
can be used in retail stores to count number of customers waiting in a queue and
if the wait is getting longer then extra billing counters could be opened to
facilitate speedier customer service. These applications are already in
commercial use, both by enterprises and government agencies not only for
security purposes but also for better management.
Video Analytics techniques
Video analytics is the process of detecting changes that occur over
successive frames of video, correlating and qualifying these changes over
multiple frames and then finally interpreting these changes. These steps can be
broken into following processes:
Segmentation: This is the process of identifying the
changes and extracting those relevant changes for further analysis and
qualification. The pixels that have changed in successive video frames are
called as “Foreground Pixels” and those that are in same place are called as
“Background Pixels”. The degree of change is used to identify the foreground
pixels and results in one or more foreground blobs, where a blob is a collection
of connected pixels.
Classification: This is the process of qualifying
each blob and categorizing it. This results in broad categorization of each blob
into distinct identifiable classes as person, car, object, etc. This requires
recognition process to be incorporated into Classification stage. After
classification each blob in the single frame is identifiable to pre-defined
object and are selected in a manner such that they provide sufficient
discrimination between each classified blob.
Tracking: The classified foreground blobs take place
over a series of successive frames as the object moved through the scene. The
position of the blob is identified in the starting frame, and the trajectory of
the object can be calculated by connecting the position of the blob in
successive frames.
Activity Recognition: This is the final step where
the results of tacking and classification are combined to correlate multiple
blobs so as to infer the activity occurring in the video. For instance two
blobs, one corresponding to a car and another to a human are merging, then it
can be inferred that the person got inside the car.
Once the foreground objects have been segmented,
classified, tracked and their activities identified. Then the behavior of other
objects in the scene relative to the given object can be analyzed. For example
incidents like a person found loitering in the unauthorized area or a bag lying
unattended in the airport lobby would trigger the system to raise alarm and
notify the security personals.
Types of Video Analytics solutions
Video analytics extends beyond motion detection and incorporates many
advanced analytics capabilities like facial recognition, behavior recognition,
people counting and wide-area perimeter intrusion detection. Vendors are
offering software suites that cater to either of these capabilities or are a
complete suite of all these capabilities. Video analytics can either be done at
the server end or at the edge, i.e. within the camera. Whereas analytics being
done on the server end can be from third-part softwares like CyeWeb from NovoSun
that takes the video feeds from IP surveillance cameras and analyzes them for
user-defined incidents.
In-Camera Analytics: Embedding video analytics
technology on the edge devices like IP cameras dramatically reduces the
bandwidth and storage requirements by eliminating the need to send video feed
across the network to a centralized server for process of analysis and storage.
By analyzing the high-quality video at the point of capture increases the
accuracy of the alarms generated by analytics, thus reducing false positives and
increasing the belief of security personals on the alerts generated. By
embedding analytics within the camera, the edge devices can analyze video feed
in real-time as it is being captured, and when a rule is violated the
notification is sent instantly to security personals, thus decreasing the
response time as well. The organizations can easily deploy and scale the video
analytics embedded devices without the need to invest on additional servers for
video analysis and on storage.
Today the IP cameras themselves combine recent chip designs
that allow better compression technologies (such as H.264) to enable high
definition video transmission with less bandwidth. This in turn has resulted in
in-camera analytics for large scale surveillance systems using megapixel
cameras. Further, advanced digital signal processing technologies and hardware
accelerators do the heavy encoding/decoding of the video to handle intense
processing requirements of video analytics within the edge device (IP camera).
Server-End Analytics: Embedded analytics in cameras
is ideal for situation where /2010/images/video is coming from one particular camera.
But when video feeds are coming from multiple cameras, as in a shopping mall,
then to perform video analytics you need higher processing power and central
management capabilities of a centralized server-based solution. Such solutions
enable more complex analytics and quicker searches of the video archives. Video
analytics that require facial recognition or object recognition or any archive
search based on a particular image match are beyond the realm of in-camera
analytics solutions. In large retail outlets, a centralized server based video
analytics solution would be beneficial since it can be used for counting people
who have entered the outlet, if any customer has not stolen any object off the
shelf, by counting people in queues it can be used for better crowd management
as well, because the video feeds would be coming from several cameras spread
across the outlet and would notify the security manager based on incidents
collected from collection of cameras and not just one particular camera.
Thus, video analytics enables security personnel to prevent
a crime from happening instead of investigating it after the incident. It also
minimizes the operational cost of a distributed video surveillance solution by
allowing real-time centralized administration and monitoring.