ipConfigure Google Search for Surveillance UIBy: John Honovich, Published on Apr 22, 2013
One of the most interesting ads at ISC West featured a search for 'computer stolen from d' with autocompletion choices of 'stolen from desk', 'data center', etc. While this is common for computer search engines, like Google, doing this with your surveillance VMS is essentially unheard of. As such, this raised a lot of questions about ipConfigure, the company claiming to deliver a 'Natural Language Search Engine' for video surveillance, as show below:
We asked ipConfigure for a short video so we could see it in action:
It sounds a little like magic, and indeed, with the examples provided, it is. An obvious question is "How does the VMS know what is a computer, laptop, office or desk?"
What's Going On
There's three main components to this search:
- The engine 'understands' what terms like 'yesterday', 'last week', 'over the weekend' mean and translates them into date / times to search. Likewise, when a user enters words like 'office', 'lobby', it converts it to camera IDs by scanning the names associated with each camera.
- Users define motion detection zones with names, like 'desk', 'laptop' etc. The engine then limits searches to those areas within those zones. This is problematic as most people do not define such detailed zones.
- If video analytics are integrated, the engine can detect relevant terms, like 'license plate KBJ-789' and use it to search for video with that matches LPR events with such a plate.
ipConfigure is building this in as a standard feature of its new 6.x VMS release. Searching for analytics requires using analytics that are compatible with ipConfigure, such as their LPR app/service.
Certainly, it has marketing appeal. Google is widely considered to be at the pinnacle of high technology plus users have grown very comfortable typing in searches using phrases. With no real surveillance VMS options, it is likely to draw further interest.
However, we think the ad is misleading. Most will not set up zones for 'desk', 'computer', etc. That said, even if it could just pull up cameras or clips for certain time frames, it could be useful (e.g., a common search like 'show me motion events in the parking lot Thursday night').
[UPDATE: ipConfigure responds: "You mention that most users do not define detailed zones or regions in their VMS, and while that's true, it's only because that until now there's been little reason to do so. Once users see and understand the power and usefulness of natural language, they'll be much more strongly incentivized to do that small amount of initial set-up work." For areas that may be common sources of searches, we can see this but for an example like a 'laptop' or 'computer' they would almost certainly have to define that first for what is likely a one time search. As such, it seems faster and easier for us to just do a traditional motion in a region search for those cases.]
Ultimately, though, the big question is how usable it will be for practical VMS searches. For example, how well does it understand the intention of requests. If it can do so regularly, it's certainly quicker and easier to type a few words than to make a number of drop down selections. That said, we have not used it so we cannot take a position, but it's certainly worth tracking.