Agent Vi's 4.0 Analytics and Auto Calibration Examined

In this note, we review the key features and positioning of Agent Vi's System 4.0 real time alerting release [link no longer available]. It includes 2 key features that are major improvements from the previous 3.3 release: (1) Automatic Scene Learning and (2) Repetitive Movement Auto Masking. Both of these address key concerns we had in our 2010 Agent Vi alerting test. At the end of the note we examine competitive positioning and contrast to VideoIQ and ioimage.

The Automatic Scene Learning functionality provides auto calibration, designed to eliminate the need for having a subject walk through the field of view while an operator configures the analytics. Agent Vi reports a few key aspects to this feature:

• The system automatically detects if a camera is indoors or outdoors (previously the user had to set)
• The system automatically detects if the camera is mounted at an angle or overhead position
• Calibrates based on humans and vehicles in the scene
• The calibration is done on the server side (the system has an 'agent' on the camera being used and a server side application)

Agent Vi estimates 30 - 90 minutes for performing the auto calibration. While the system is calibrating, alarming is disabled. Agent Vi says that the calibration does not need to be performed each time the camera is rebooted as the calibration configuration is stored on the server.

The Repetive Movement Auto Masking aims to eliminate problems from swaying tree branches, water and other 'noise'. Agent Vi describes this as "based on continuous statistical analysis of movement in various areas of the image and learning of what is “normal” movement in a specific area. It takes about 2-4 minutes to adjust to new areas of repetitive movement. What is exceptional about this feature is that it can still detect moving objects against the background of repetitive movement that is being filtered." Like many analytics we have tested, the previous version of Agent Vi (3.3) experienced significant problems with swaying tree branches.

From a usage standpoint, a few points to note:

• The new 4.0 version is commercially available
• Version 3.3 can be upgraded to 4.0 for users with a Software Ugprade Plan
• Overall camera computing usage increase is estimated to be ~10%
• The 4.0 server can support cameras running 3.3 and 4.0 agents (backwards compatability for cameras)
• No change in pricing / licensing cost

Additionally, Agent Vi has released a new version of their search analytics. For details, see the Agent Vi Search Version 1.2 release notes [link no longer available].

Competitive Positioning

We think these 2 new features are important advances for Agent Vi as they address the two most significant weaknesses of the previous version - setup complexity and false alarms for 'noisy' scenes.

Historically, Agent Vi's biggest strength has likely been its ability to run on different manufacturer's cameras (most notably Axis). This allows using existing IP cameras and the specification of 'regular' IP cameras. By contrast, two of the most common 'smart' cameras used (VideoIQ and ioimage) both require the use of their own IP cameras (note: both have encoders for analog cameras).

If Agent Vi can match VideoIQ's auto-calibration performance and reduce their repetitive motion false alerts, Agent Vi could become significantly more attractive. The combination of that plus their support for multiple camera vendors would be a strong combination against VideoIQ. On the other hand, VideoIQ offers analytics across megapixel resolution while Agent Vi is limited to max SD.

We have not tested the new Agent Vi release. However, we do plan a new round of video analytic testing this spring.