BRS Labs AISight 3.0 ExaminedBy: Ethan Ace, Published on Oct 05, 2011
BRS Labs' AISight is, without a doubt, one of the most hyped products in the industry led by one of the most aggressive marketing campaigns ever. For example, just last month, they held a press conference announcing the death of video analytics. That noted, great unclarity continues to exist about what the product is and how it is positioned relative to 'dead' video analytics. In this update, we overview what AISight actually does and does not do, and help users understand their offering.
Overview of AISight
BRS Labs is claiming something that no one else commercially is claiming - that it can identify anomalies in an environment without any set specifications and alert operators to potential risk. Even 3 years after launching BRS Labs remains the only commercial surveillance company claiming this ability.
They are not doing rules-based analytics as is typical in the analytics market. Where rules-based systems operate based on hard and fast criteria set by the user, AISight "learns" what is normal behavior for objects in a scene over time, and alerts an operator to off-normal behavior. It does not classify this behavior as a line being cross, an object removed or left behind, etc. It simply alerts an operator that something outside of normal is happening. These alerts can be different for different times of day or week, but typically this is not needed.
AISight performs this analysis by watching a scene (initial learning reporetedly takes about three weeks) and recording data about these objects as they travel through the scene, including: where objects normally enter and exit the scene, the path they normal take, dwell time, speed, and more. AISight does detect whether an object is a vehicle, a human, an animal, etc., but this is not used for alerting. Instead, it is compared to the normal paths that object type takes through the scene.
Below is an image showing side by side a camera's FoV and a visual representation of what AiSight is tracking/cataloging - that is, the normal flow of vehicles on a road:
Unlike rules based analytics, it does not pin point certain types of activity by rule (i.e., someone crossed the road at the upper left) but identifies activities that are uncommon relative to the historic patterns in the scene (i.e., an explosion in the lower corner where activitity is historically scarce).
AISight requires a server separate from the VMS server, on which it analyzes a second stream from each camera at CIF resolution. For extremely large areas, 4CIF may be used, but BRS Labs says this is normally not needed. Each AISight server (assuming their normal configuration, equivalent to an enterprise class HP DL160 with X CPU and Y RAM) can analyze about 20-25 CIF streams, with 4CIF reducing that proportionately.
Additionally, BRS Labs does not run on any cameras and has no 'smart camera' offering.
BRS Labs has limited VMS integrations completed. BRS Labs reports initial integrations with high-end PSIM and security management offerings such as Boeing, Cisco's PSOM, and Raytheon for large scale projects. BRS Labs says that integrations to VMS systems are currently in progress but not complete, including Genetec, Milestone/OnSSI, and Verint. The level of integration varies, but typically is accomplished via XML metadata, used to create alarms on the receiving system.
Additionally, of course, BRS Labs has its own client that can display alerts and related information.
Pricing for AISight is relatively expensive. Quoted MSRP pricing is $3,295 for a single camera license. However, BRS Labs says they have little interest in small camera count systems, instead seeking large projects (300-500+ cameras), at least in its initial deployments. Users seeking these quantities of channels of AISight analytics should suspect to pay less than $3,295 depending on quantity. We would estimate 30-50% discounts would be possible. In addition to the software license cost, we estimate that each server (HP ProLiant DL160 or similar) required for an AISight implementation will cost ~$6,000 - about $200-$300 per channel for hardware in addition to software licensing.
In addition to the channel license cost, BRS Labs also offers an software maintenance agreement. This is included in the first year, and includes tech support and version upgrades. Users not selecting the maintenance agreement will not lose current functionality, but will be charged for tech support and version upgrades.
Compared to rules-based analytics, BRS is, without a doubt, expensive. Most 'professional' analytic software costs $1,000 per channel or lower. For instance, users could purchase a VideoIQ HD camera with analytics built-in for less than an AISight license. BRS claims that some of this cost is offset since their system does not require calibration and subsequent recalibration if conditions on site change, where rules-based systems normally do. That noted, for users looking to monitor specific types of events - someone crossing a fenceline or being present in an areas only after hours, traditional rules based analytics are likely to be cheaper and better than an abnormal detection system like BRS.
That being said, rules-based analytics do not offer some of the functionality that AISight claims. Crowded, busy scenes have historically been a problem for analytics. BRS Labs claims to perform best in busy scenes, since they are only looking for exceptions, not for all activity. This has clear potenetial though we have not tested it ourselves and cannot validate.
We feel that BRS has an interesting offering. If they can succeed in environments where other analytics fail, they could gain traction. However, pricing will remain a barrier to widespread adoption. Typical customers in the 100 cameras and less markets (the majority of the industry) would find this pricing hard to swallow, essentially restricting BRS to the high-end of the industry. We imagine in the future, as more systems are deployed, they would seek to reduce pricing to gain market share, but this remains to be seen.