You've got cameras all over your facility. There are cameras in reception, the loading bay, the hallways, the data center; pretty much everywhere, and you accumulate footage by the gigabyte every day.
So how do you find the bits containing people? Many video systems let you set up movement triggers to flag events. These usually operate by software monitoring all or part of the video frame for changes in pixel data that exceed a certain threshold. This threshold is intended to eliminate false positives from, for example, curtains blowing in the wind or dogs wandering about.
The problem with this approach is it's crude. While slight movements of the curtains or Matilda's Chihuahua are handled just fine, a really windy day or the boss' Great Dane will trigger more alerts than you can shake a nightstick at.
Today's helping of high-tech grooviness, Vitamin D from Vitamin D, does for video monitoring what "American Idol" does for singing, to wit, take thousands of boring and irrelevant examples and show you just the good stuff.
Vitamin D was developed using the Vision Toolkit based on Numenta's Hierarchical Temporal Memory (HTM) technology. Numenta was co-founded by Jeff Hawkins, the founder of Palm Computing and Handspring, and the HTM system is based on a Hawkins' theory of how the human brain performs pattern recognition.
If you've had any experience with pattern recognition technologies such as neural networks, HTM will seem familiar. Like neural networks, HTM systems use interlinked layers of processing, but instead of emulating neurons as neural networks do, HTMs emulate the structural organization of brains.
The Wikipedia entry for Hierarchical Temporal Memory explains: "An HTM comprises a collection of nodes that are arranged in a pyramid-shaped hierarchy. [An] individual [node] of the hierarchy self-discovers an array of causes in the input patterns it receives, both in spatial and temporal terms."
In other words, this ain't yer father's machine larnin'.
What the Vitamin D product does is monitor video streams from either USB or IP cameras and, through its pre-trained HTM subsystem, detect people as well as things that are not people but moving.
Depending on rules you can define for individual cameras, when Vitamin D finds a person or an unknown moving object in the scene (optionally restricted to when the person or object enters through a door, crosses a boundary or enters a specific area of the scene), it can generate an audio alert, save the video clip with a line drawn around the detected object, or send an e-mail message with a still of what was detected.
In practice, the performance and accuracy of Vitamin D is very impressive and it speeds up reviewing video surveillance footage by orders of magnitude.
This is the first release so there are a few rough edges: For example, the software really needs a "throttle" on how many and how often notifications are sent (me walking into my office generated more than 30 notifications in just a few seconds). It also needs an API because you'll probably want to send event notifications using other services (such as instant messaging) or integrate with a network management or security system. The company tells me these features and many others will be available in near future releases.
Vitamin D is available for free to use with a single camera at QVGA (320x240) resolution, while two-camera support at either QVGA or VGA (640x480) resolution costs US$49. A version supporting unlimited cameras (typically two per processor core) and both resolutions is priced at US$199. Vitamin D is very impressive and amazingly cost effective compared with other available solutions.