|Video Analytics versus IPVS 2013|
This article attempts to explain what Video Analytics (VA) is and how it differs from IP video surveillance (IPVS).
VA is about “analysis of images” captured by surveillance cameras of an IPVS system in real time. IPVS analyses images or camera footage in order to produce useful information for surveillance purposes. For example, we can get alarm alerts from an IPVS system when motion is detected in a pre-defined area by one of its cameras. VA does the same and a lot more using computing resources outside of the camera and IPVS system.
Motion Detection works on change of pixel state (from 0 to 1 or vice versa). The user pre-defines how many pixel changes constitute motion. This algorithm is simple and can easily be done by the camera or IPVS system. The trouble is that pixel changes occur in many situations such as environmental lighting change during day and night, falling leaves, cats and dogs, tree shades etc. Many alarms produced by an IPVS system are not what the user needs and they have become nuisances. VA is here to reduce or even eliminate nuisances.
In addition to security surveillance of the area under monitoring by cameras, the same image captured can be analysed to produce traffic counts, vehicle license plate number recognition, face recognition, abandoned objects, violation of traffic regulations, and so on. Applications of VA are indeed limited by our imagination.
(Hover mouse over to enlarge)
To be meaningful, most VA applications have to produce information or alerts in real time. This condition requires high performance computing of camera images. Therefore VA is normally done by a separate PC to the NVR of the IPVS system. Camera footage is fed to both VA and IPVS for processing. Unfortunately VA requires different view perspective to IPVS in some applications such as people counting versus intruder recognition. VA is interested in heads and IPVS is interested in face, dress and body shape. That is, VA may need dedicated cameras to be installed.
VA is still not a common product off the shelf as of March 2013. Whilst many IPVS systems offer motion detection of up to 16 cameras for free, one VA license for the recognition of licence plate number of vehicles has a price tag of about $5000 for car park policing and $10,000 for motorway policing. Applications with a higher level of intelligence will be more expensive as a rule of thumb.
One camera needs one license per application. Several cameras can share the same VA server as it is a computing resource issue. In case the VA function is to track a person across several cameras, those cameras have to go the same VA server.
IPVS will incorporate more functions that are current in the domain of VA. This is a natural technology progression issue. For sure, VA will move on to tackle more complex scenarios.