iCetana Video AnalyticsBy IPVM Team, Published Nov 10, 2014, 12:00am EST
However, BRS Labs' core claim of behavorial recognition does have value if it can be delivered.
Indeed, a quite company on the other side of the world says they can do very similar things.
We talked to them to learn more about their offering, its positioning and pricing.
While video analytics market is full of aggressive boasting and overpromising, Australian developer iCetana resists being cast into the same mold, and instead is reserved, low-key, and bluntly publishes system limitations on their webpage.
"iCetana technology is based on a patented innovation we call iMotionFocus. This technology does not try to perform any background/foreground separation, object identification, nor rely on user-defined specifications. Instead, it automatically learns the “normal” motion patterns of people, vehicles and objects in a scene."
The keys for being a good fit for iCetana's analytics? The company says video systems where manned operators are staffed to ultimately judge 'security events', where the system employs more than 50 cameras, and where typical surveillance activity falls into patterns.
Part of the 'low-key' approach is the company is currently transitioning from an early spin-off, into a company engaged in selling merchantable product. The company has been focused on expanding and optimizing existing systems in Australia, but plans to ramp up sales efforts outside its home market in the months ahead.
The platform is based on 'exception events', not object detection. Regardless of native camera resolution, the entire camera Field of View is downscaled to a 4CIF, 5 FPS stream, where iCetana's software compares pixel changes in the current scene to a typical template of the way those pixels appear.
The 4CIF image (704X480) are broken into 400 subzones that can be individually tuned to alarm on different sensitivities. These analytic alarms are based on comparing pixel color, pixel change direction, and velocity of changes against a 'normal' condition. Recalibration of the model scene happens first after 14 days, and then daily afterward, which the company claims minimizes false alarms due to periodic changes like weather or foliage differences.
The company claims that analyzing pixels, not detecting objects, results in a faster analysis and fewer false alarms.
Operators Still Needed
The primary claim iCetana makes is that it reduces the amount of video events an operator must view, so that only highly irregular or atypical events are shown. iCetana claims that it alarms on less that 1% of typical video but that even then, a slim percentage of those prove to be 'actionable security events'. However the claimed benefit is filtering out the need for video operators to needlessly review hundreds of hours of 'normal' unimportant video.
At this point, iCetana is only selling to certain markets, and an active central video monitoring operation is required for deployment. The company is selective on which systems it will install product into, attempting to avoid system failures and poor results.
However, unlike the claims of other VA vendors, it cannot pick faces out of a crowd. Even normal VA uses like 'people detection' can be a challenge with a 4CIF stream. Essentially detecting human movement further out than 50 feet for a typical camera in a well lit scene depends on how crowded or busy the scene normally is. The company explained performance in these terms:
"We detect objects/events that operator can see reasonably clearly. So small things in background that are hard for operator to see/concentrate on will also challenge us. [Also] very narrow field of view cameras (e.g. framing single doors) are ok for in/out of door, but not for cross view traffic which can cross very rapidly, where there are not many frames worth of data."
iCetana is installed on standalone servers sold as appliances. The company claims that much of the actual analytics is GPU based, and a single iCetana server can handle a maximum of 100 cameras.
The iCetana server also records events in a revolving library from the previous 30 days. Even if video is overwritten or deleted in the main recording VMS, the analytics appliance stores a redundant copy of the alert.
The company sells hardware and software as a single part number, and all deployments are sold the same basic server and analytic package in groups of 100 cameras per server.
Rough MSRP pricing falls between $1000 - $1500 per channel, for a minimum appliance/software/service cost of about $100,000 - $150,000. In year two, the maintenance option costs about 15% of the total, or about $15,000 per server per year. The company lumps hardware and software support into the ongoing maintenance fee.
The company essentially self-installs every analytics deployment by sending their own engineers to work with partner resellers and integrators.
iCetana's lists its closest VMS parters as Milestone, Indiovision, and DVTel, based on previous project integrations. However, the integration between iCetana and VMSes are basic, comprised of input alarms that index events captured on iCetana's separate server.
The company started in 2010, as a spinoff of a Curtin University computer vision research project. To date, the company has deployed to municipal buildings, transportation hubs, and universities within Australia, but plans to execute its first North American project in Q4 2014.
1 report cite this report:
Back to Top