Video Analytic Search from AgentVi Examined (Vi-Search)

Published Sep 20, 2009 00:00 AM
PUBLIC - This article does not require an IPVM subscription. Feel free to share.

Using video analytics to enhance search and investigation is growing. AgentVi has announced Vi-Search [link no longer available], a new product offering that complements its existing alerting solution, Vi-System [link no longer available].

Search Offerings

This joins a small but growing handful of companies offering search, including 3VR who focuses its marketing on 'searchable surveillance' as well as VideoIQ and Bosch who both have built-in search functionalities in their video analytic offerings.

Overview of Vi-Search

AgentVi Vi-Search [link no longer available] provides a number of advanced functionalities (including search by color and person as well as statistical analysis, motion path analysis and video summary) using the same fundamental architecture as their alerting product. (read the Vi-Search brochure [link no longer available]).

One significant differentiator is that AgentVi is highly focused on working with 3rd party VMS systems as AgentVi does not sell their own VMS system (unlike the other manufacturers mentioned above).

Questions

While using video analytics for search is generally considered 'easier' than alerting, because of the value of stopping attacks, the emphasis for years has been on alerting. 

Vi-Search will be deployed similar to AgentVi's alerting offering. An agent will be loaded onto 3rd party cameras like Axis, IQinVision, Sony, Verint, Vivotek, etc. The first part of the processing will be performed on the camera. The second and final part of the processing will be performed on a centralized server.

Both alerting and search functionality can be run on a camera. On the server side, up to 200 cameras may be processed on a quad core server with 4GB of RAM. Alerting and search consume similar levels of computing resources. If alerting and search is both run for each camera, the maximum cameras on the benchmarks above is 100 cameras.

For the search functionality, AgentVi generates meta-data including object characteristics and timing. When an operator conducts a search, video is called from 3rd party VMS systems (currently including Milestone, OnSSI and Vicon). AgentVi's Vi-Search does not record video itself.

AgentVi will be providing its own client UI (see images below for examples). However, AgentVi aims for 3rd party support inside of VMS clients to make it easier for operators to use in their main interface. Support inside of the VMS UIs is anticipated in the middle of 2010 but is not yet committed.

Vi-Search offers a range of functionalities:

  • Searching by person or object with filters for adults vs children and small/medium/large objects.
  • Persons or objects can be filtered by types of behavior including moving, stationary, crossing lines, occupancy, etc.
  • A sensitivity filter is provided. Searches can miss valid matches as well as return false matches. The sensitivity filter allows investigators to choose whether they want more valid matches (at the expense of more false matches) or less false matches (at the expense of less valid matches).

These basic functionalities are displayed in the image below (specifically, the top row):

The object path functionality is more advanced. Essentially, graphs are displayed representing each object's path. An individual graph may be clicked upon to replay the video associated with it. Operators can specify time frames to generate these object paths. Compared to generating galleries of thumbnails, this can make it easier to quickly analyze a broader time frame. A sample image of object path functionality is displayed below:

Additionally, statistics reports and video summary functionality is included. The video summary function is quite similar to Briefcam's synopsis tool.

Pricing

Pricing for the software license is $200 MSRP per channel in addition to a per server fee. Vi-Search is scheduled for production release in December 2009. 

Performance

Since Vi-Search is not yet released, it's not possible to evaluate performance. 

Competitive Positioning

Vi-Search faces 2 competitive challenges:

  • "Standard Search": the status quo today is to use the built-in, basic search functionalities built-in to VMS systems such as time based and motion detection search; while not complex, these functionalities address the most common use cases and is generally offered 'free' as part of the basic package
  • Other specialists in video analytics: organizations may consider AgentVi vs. 3VR, Bosch, VideoIQ, etc for advanced search. However, it is unlikely that this will be done in isolation since these offerings require the use of specific VMSes (i.e., if you want to use 3VR's search functionalities, then you have to use 3VR as the VMS, same basically for VideoIQ and Bosch). However, because competitors bundle in search with their VMS or cameras, search is generally offered at no (or little) additional cost.

One of AgentVi's strongest structural advantages is its positioning search as an open 'plug-in' to leading VMS systems (such as Milestone and OnSSI). This creates an opportunity to work with many VMS systems and a larger overall market.

However, there are 2 important downsides to this approach. As a 3rd party plug-in, Vi-Search requires integration into other systems and additional cost. If and until Vi-Search can be accessed from inside VMS client UIs, using search will require extra steps and extra user software. Such integration is complex and may require fundamental changes to the UIs VMS use. More importantly, it will be difficult for Agent Vi to match the price of competitors who bundle in advanced search with other products they sell.

Another advantage AgentVi has is the more advanced options provided such as the object graph and the video synopsis. These features are not provided in current products by 3VR, VideoIQ nor Bosch. If AgentVi can maintain an edge in more advanced search functionalities, this could increase their competitive value.

Best Fits

Given the incremental cost involved and the limitation of VMS Client UI integration, customers may need to have a high number of ongoing investigations to justify purchase of Vi-Search.

From an ROI perspective, two key elements are:

  • How many investigations per cameras does an organization have per year? An organization may conduct 100 investigations per year but it makes a big difference if it's across 1,000 cameras or 10,000 cameras. Specifically, this is important because of the per-camera license fee. If an organizations has very few investigations per camera, the licensing could be very expensive relative to the potential savings.
  • How many additional investigations can be solved per year? This could be existing investigations or investigations that were too difficult to be attempted previously. Ultimately, this will be a key factor in justifying the use of these analytics.

Organizations would be prudent to test/deploy analytics on a small number of cameras to start that cover the most frequent areas requiring investigations.