Ipsotek Analytics ExaminedBy Ethan Ace, Published on Nov 20, 2011
The video analytics industry has nursed a black eye for years, thanks to overhyped marketing claims and inaccurate technology. At the same time, a handful of vendors have begun to come forth, however, with claims of reduced or eliminated nuisance alarms and increased accuracy. British analytic provider Ipsotek claims to do so with a hybrid of cognitive / behavioral analytics combined with traditional rules-based analytics. In this note, we look at Ipsotek's approach and how they compare to others such as mega-hyped cognitive analytic provider BRS Labs.
Ipsotek offers a suite of analytics which may be combined. They refer to as "scenario-based" analytics. The Visual Intelligence Suite (VISuite) contains multiple analytic rules, such as the direction of travel, virtual tripwires, loitering, etc. Ipsotek allows these rules to be used simultaneously or in sequence, specific to the scene, in order to more precisely create alarm scenarios. An abandoned car may be detected, they claim, not simply by a car stopping, but by a car stopping, two subjects getting out, and not entering the nearby building, but leaving the field of view elsewhere. These analytics scenarios are created via a drag and drop graphical interface.
In addition to the camera, Ipsotek also allows external systems or sensors to be used as part of the scenario. Any sensor offering a relay output may be attached to camera inputs so their state may be detected. VISuite is also capable of receiving XML streams from third-party systems, such as LPR or access control, to use as a trigger in a scenario rule. This integration with sensors allows a fence detection sensor or buried cable to start the rule, for example, but only create an alarm if a person is present on camera, or taking a specified path.
While the basis of Ipsotek's offering is rules-based, the system also uses cognitive analytics to reduce the number of false alarms presented to the operator. This is not the same as BRS Labs' behavioral analytics, which learn and present off-normal behavior to the operator. Ipsotek claims that over time, as users mark alarms as false, VISuite reduces falses alarms from future similar events. However, as cognitive analytics systems must learn behavior over time, they must learn from operator feedback over time. Simple rules-based systems can hit the ground running, but are more prone to false alerts. This combination is what Ipsotek claims provides the best of both technologies.
Tag & Track
Ipsotek has also created a second-generation product, named Tag & Track, which is intended to be used for tracking people and vehicles across cameras, either in real-time or in forensic search. Ipsotek claims this can be done automatically without predefining relationships among cameras. This is the first we have heard of such a claim being in a production analytics system. The only thing that sounds similar is the IBM system though we believe operators still track manually across cameras.
Ipsotek says they store metadata about an object, such as its colors, size, speed, direction, etc., and using this data to locate the same object on other cameras. Cameras also "learn" where they are in relation to each other based on the frequency of objects appearing between them. For example, if objects frequently enter on the left of camera 1, exit on the bottom right, and appear on the top right of camera 2 ten seconds later, this is considered a "normal" path. This data is used in conjunction with object metadata to assist in tracking objects across cameras. The object's path is presented to the user graphically, via a map view, along with the video of the subject as it crosses each camera.
Ipsotek admits that Tag & Track has limitations. When a scene is particularly crowded, such as a busy airport or special event, it becomes very difficult to differentiate and track individual subjects. It may function, but produce a lot of inaccurate "hits", so an operator will have to manually filter through them.
Though it is currently available, Tag & Track is currently in use in only one deployment, which was started as a pilot. Pricing for Tag & Track is unavailable, as it varies widely depending on each user's requirements and the size of the system.
Ipsotek analytics are designed to run on their own hardware, which is available in two form factors. The VIS200 is a 2U 19" rackmount server which can run up to 12 channels of analytics, either analog cameras directly attached to DSP cards, or IP streams. The VIS1000 is a single or dual-channel IP65-rated edge unit. Designed to be located near the camera, the VIS1000 contains a fanless PC and solid state drive, and may be connected via on-board wired or wireless Ethernet, or 3G modem. Users are also able to run VISuite on their own hardware, but Ipsotek's DSP cards must be installed in the server, as the line was designed specifically to use these DSPs. As such, Ipsotek is also not designed to run directly on cameras.
Ipsotek reports integration with a handful of VMSs: Genetec, Milestone, Indigo Vision, Wavestore, and BAE Systems UVMS. Additional integrations are created on a project basis. Ipsotek is capable of sending metadata alerts to the VMS, though it stores this data in its own database.
Pricing per channel for VISuite varies based on quantity, starting at £2,000 (~$3,200) for low camera counts, down to about £700 (~$1,100) for large deployments. The low side of this pricing is roughly in the range of other analytic competitors, such as AgentVI. The high side is nearing BRS Labs' pricing, which starts at about $3,500 per channel. Currently, Ipsotek, based in the UK, has little North American representation.
In the analytic field, there are a lot of small competitors, most of whom are of the rules-based variety. We know of no other analytic providers offering the combination of rules-based and cognitive analytics. Against typical rules-based competitors, Ipsotek may provide better nuisance alarm rejection, though it may be priced above other options on smaller camera counts. Against BRS Labs, it offers the deployment advantage of rules-based analytics, which alarm as soon as they are deployed, while BRS Labs takes time to learn the scene.
However, compared to other rules-based options, Ipsotek is relatively expensive. Additionally, they have no smart camera offering, such as competitors like VideoIQ or ioimage, making them costlier for small camera counts. North American users should also take into account Ipsotek's lack of presence on the continent. With sales and technical support both originating in England, response times may be higher.