Thursday, May 14, 2009

Forensic search

All video content analysis solutions tell you that something bad is happening right now. That's a given, and some solutions do it more reliably than others depending on the scene, and they also vary in their detection capabilities.

But one of the great strengths of the Bosch Intelligent Video Analysis (IVA), like a few others, is not just its ability to generate live alarms, but to search back through days, weeks or longer of video, hunting for that critical event - that needle in the haystack. In the simpler world of motion detection the analogy is smart motion search, also known by other names. With SMS you can highlight the area of a parking space and ask it to search for whenever motion happened there. Very simple and effective with the well known limitations of pure motion detection. The same concept is true with IVA, but it's just taken to a new level. With Forensic Search you might want to search through 6 months of video and ask it 'Tell me all the times someone parked in front of this gate' in order to find out if it is a common occurrence. Or to find out when the graffiti was artistically applied 'When in the past week have people been loitering in front of this wall for more than 2 minutes'.

Equally important to searching after-the-fact, it is also a critical set up tool. Set up the camera and record it for a few hours, and use your friends to reluctantly simulate the kinds of stunts you want to detect and ignore, but don't set up any rules just yet. Then using forensic search define the rules and test them against that recording. It will instantly tell you which events caused alarms and which ones were ignored. Tweak the rules until you get the performance you expect. The great thing is you don't have to go back time after time to re-enact those scenes, because they're all already recorded and analyzed. Consider this a life saving tip for saving time during IVA setup. And you will keep your friends for longer.

As a final note, some people understandably can't get their heads around 'metadata'. Video content analysis information, in the form of metadata, is generated and stored with the video. The recorded metadata, comprising simple text strings describing specific image details, is much smaller and easier to search through than the recorded video. The metadata includes many things including an objects size, aspect ratio, speed, location and direction and color. The kinds of things that help you to draw a new trip wire and ask 'Has anything, over the size of a small bunny rabbit, crossed this line in the past month?'

For background reading, a brief but good step by step article is Campus Safety and Security's 8 Factors to Consider When Deploying Video Analytics by Craig Chambers. It raises many great points, as valid now as it was back then

Steve Hunt, one of the industry's premier movers and shakers, at least on my Richter scale, paints a complete primer in his podcast at

Another good set of warnings are presented as objectively as always by John Honovich of fame at

Monday, May 11, 2009

Quick tour of ISC

This brief movie doesn't replace flying to Vegas for ISC to meet us at the booth, but it takes less time than it takes to pack your favorite party shirts for the evenings.

Take a whirlwind tour of cameras we baked, drowned, froze and pulverized with dirt, sand and fine dust. I need to dig up the video where someone was beating the living daylights out of another camera with a hammer. I found it ironic that the camera survived the 3 days of continuous beating, but that the table actually broke. And yes, the hammer was OK too. Finally check out the fully automated License Plate Recognition solution (including the DiBos DVR, Access Control and the barrier) that was reading Bond-like revolving plates on a Mini with the camera overloaded with full headlights.

Courtesy SP&T News (Security Products and Technology News) and my Canadian heroes, Norm Hoefler and Willem Ryan. Click here to watch the movie,com_seyret/Itemid,119/id,10/task,videodirectlink/.