Seems parts are a bit misleading and exaggerating capabilities. E.g.
>> Images from the cameras are in turn fed to computer programs that allow analysts to track suspects, and even to rewind to look back over their paths, like watching TiVo.
I'm pretty sure that really just means that they digitized footage and the analysts did the actual tracking manually but it reads as having good image recognition back in 2006.
> I'm pretty sure that really just means that they digitized footage and the analysts did the actual tracking manually but it reads as having good image recognition back in 2006.
I do not know the specifics of the software used in the Gorgon Stare program, but SRI started work on automated image surveillance under DARPA contract in 1982 with ImagCalc. The system was later expanded to video and continued development until two years ago:
I am not sure what year they added automated video tracking. At the 2007 International Lisp Conference Christopher Connolly and/or Lynn Quam (can't remember) showed a demo of FREEDIUS that, among other things, had automated track analysis on aerial and CCTV surveillance footage; by that time the problem was long solved, and they were working on automated event detection.
Same year (2007) the same SRI group also published this paper, "Recovering Social Networks From Massive Track Datasets":
I'm not sure it depends on image recognition as in visual classification (although a lot of that work was DoD funded). Once you identify a target shape/pattern/object, you can use motion tracking to track it backwards in digitized video.
>> Images from the cameras are in turn fed to computer programs that allow analysts to track suspects, and even to rewind to look back over their paths, like watching TiVo.
I'm pretty sure that really just means that they digitized footage and the analysts did the actual tracking manually but it reads as having good image recognition back in 2006.