A key factor distinguishing Cobalt from other video detection systems is that it attempts to store all relevant data about a scene along with a compressed version of the scene video. What is meant by "relevant" in this context is "any change of significance that might potentially be the source of a video query". So it tries, for example, to ignore changes in illumination, persistent foreground or background interference, or moderate camera shake. Another way of putting this is to say that Cobalt seeks to encapsulate "what is going on where in the scene". The encoding of this stream of information we refer to as "Image-DNA".
The Image DNA data is delivered as a separate low bandwidth data stream, alongside the compressed video data. It is not image meta-data, though image meta-data might be constructed from it. Image-DNA streams are searchable in response to queries relating to the question "what, if anything, is happening or happened at these locations?". Such queries are formulated graphically rather than through the more familiar text-based search, and the output of such a search could indeed be saved as "event meta-data", though it is usually presented visually as an event replay.
Because the Image-DNA is highly compact the response to event queries is ultra fast: an hour of video data can be scanned in under a second, and a day in around 30 seconds. A month of data is scanned in 15 minutes. The ability of the Cobalt system to adapt in real time to ambient conditions means that a single query can span day and night and most variations in weather conditions.
Cobalt is an engine that forms a basis for building diverse video analytics systems. These systems fall into two classes: real-time and post-recording analysis.
Real-time systems are generally thought of as security systems that monitor some space, indoors or outdoors. Initially, this was Cobalt's primary target. However, Cobalt has been used in industrial process control, vehicle parking and other situations, often outdoors.
Post-recording analysis of video has mostly been used in the security and surveillance sectors. The ability to search quickly and easily for events occuring over long periods of time has saved enormous amounts of human effort.