Video Analytic False Claims
Numerous wildly inaccurate claims on video analytics were made in today's Dallas News [link no longer available].
The negative impact of such fictional claims is a threat to security and our industry.
Here are 3 of the worst claims from the article:
1) "These systems will reduce crime by several orders of magnitude," said Dr. Bruce Flinchbaugh, director of TI's Video & Image Processing Laboratory.
Reducing something by several order of magnitudes means a decrease of 90% to 99.9%. Does anyone, outside of Dr. Flinchbaugh evidently, really believe that video analytics are going to essentially wipe out crime? Perhaps he was quoted out of context?
2) "The way a tourist circles the base and gazes up at the tower is different than the way a would-be climber does. The tourist is relaxed. The climber is tense," said [NICE VP of Sales] Mr. Teegardin, whose company works closely with TI. "We have taught the program what to look for."
Can NICE's analytics really differentiate between a person who is relaxed and a person who is tense? Can NICE really differentiate between how people glance? I suspect that what this analytic does is monitor some form of loitering (taking this quote literally what they are describing is a research project, not commercial technology). If it is loitering, that could be valuable for one of the world's greatest landmarks. At the same time, the false alert rate must be very high. Such a rate may be acceptable given the security needs and budget for the Eiffel Tower but certainly not to the school systems that might want to use for similar purposes.
3) All of Olympic Centennial Park in Atlanta could have been evacuated long before the pipe bomb exploded in 1996.
Think about how hard it would be to even get the camera coverage necessary to cover an area like the Olympic Centennial Park. It's a very large outdoor space so you can't simply mount camera to the roof. You will have to mount them on poles, lots of them, fixed cameras too. Covering more than a fraction of such a facility is not feasible.
Even with that fraction, abandoned object detection is one of the hardest analytics to have work accurately, especially given the dense activity in such locations and the need to trigger an alert very quickly.
As an industry, we must have better and more honest communication. This is perhaps the most important objective I have for this site. It is just too hard to find accurate information and it is to the detriment of all of us.