Video analytics is the thing right now. With so much video information and the limits of human operators, what to do with all that footage? There are numerous answers mostly involving forms of algorithmic movement recognition. From the masses of press releases that come my way every day, I notice one Israeli company (Israel being one of the world leaders in security technology), BriefCam, is marketing a new automated system that not only recognises objects of interest but then condenses hours of video which feature the object of interest into a matter of minutes featuring all the salient points from the whole time, at once. Or according to their own website:
“BriefCam VS Online receives real-time feed from the DVR/NVR and processes data to produce a database that can be called on to create a video summary presentation, on demand.”
I’ve seen the technology at work, but one thing starts to concern me imediately is what is lost by way of this combined footage. Check the video here for example.
The blurb claims that it is ‘easy’ for operators to see something unexpected, yet this is not a ‘real’ image, or in fact it is a hyperreal image, multiple images partly overlaid on what is assumed to be a standardized background. Of course, given the original footage remains available contextual evidence can be sought. However, I do wonder what kind of decisions will result from fast-moving combined footage pre-selected to present to a human viewer… and of course, what exactly it is that the system is programed to recognise and how. It seems that operators of video surveillance systems will increasingly be watching is not reality, but combined, edited highlights, a part-simulated recreation. Jean Baudrillard would be having a quiet chuckle if he could see this…
There are some reports circulating around the web that researchers from the Technical University of Ilmenau, Germany, have invented an algorithm for unobtrusively erasing objects from live digital surveillance camera footage. Now the possibility of post-hoc manipulation of video has long been known, but the idea that live images could be altered is something new. A device that could trigger such an erasure drove the plot of the superb surveillance technothriller, Whole Wide World, written by Scottish author, Paul McAuley back in 2001, but almost ten years later, reality appears to have caught up with a piece of near-future SF that already felt perilously close.
According to Ray Kurzweil’s blog, the software is being demonstrated as I write at the Symposium on Mixed and Augmented Reality (ISMAR) in Seoul, although the researchers appear to refer to their invention as ‘diminished reality’. There are links to video on the invention from both there and the university press release (above). The software appears to work by recognised shapes and removing them from the video as the feed comes in and before it reaches any display.
However, neither Kurzweil nor any of the other commenters on this story (e.g. BoingBoing) seem to get the potential seriousness of this development, both for resistance to surveillance and for the credibility of video surveillance: it could be a fantastic tool for privacy, or an equally fantastic tool for social and political control. It’s one thing to be able to manipulate the past (to do what Stalin did to his oppenents and airbrush them out of history -see David King’s excellent book, The Commissar Vanishes), it’s yet another thing to be unsure whether what one is watching on TV or on YouTube is ‘real’ or ‘fake’ or some combination, but it is another thing entirely to be unsure whether the supposedly live images from a surveillance camera are actually real or not…
The Gothamist blog has a brief report on the massive upgrading and expansion of the video surveillance system in the New York public transit system. Like Chicago, which I’ve mentioned several times here, the cameras in New York are really just collection devices to feed an evolving suite of video analytic software, that can track suspects or vehicles in real-time or search through old footage to find multiple occurences of particular distinctive objects or people.
The other notable thing is that the new camera system is just completely overlaying the old – in other words there is no attempt to connect the older cameras which are not compatible and have far poorer image quality. As cameras and software gets cheaper, this option looks like being the one many urban authorities will pursue, so cities like London, which pioneered widespread video surveillance, but which, with their disconnected mosaic of incompatible systems, have started to look increasingly ineffective and out-of-date, could deal with this not by expensive and unreliable fixes but simply by sticking in an entirely new integrated algorithmic system on top of or alongside the old ones. Technological fallibility and incompatibility can no longer be relied on as protections for the privacy rights of citizens in public spaces.