FAILED: Facial Recognition & Boston BombingAuthor: John Honovich, Published on Apr 20, 2013
Facial recognition vendors have been tripping over themselves to take credit, insert or claim the wonders of their offerings to identify the Boston bombers. However, an excellent Washington Post investigative report cites Boston's Police Commissioner, confirming that:
"Facial-recognition software did not identify the men in the ball caps. The technology came up empty even though both Tsarnaevs’ images exist in official databases."
So what solved the case? A lot of:
"Work [that] was painstaking and mind-numbing: One agent watched the same segment of video 400 times."
Plus, a critical tip from a victim who had seen the bomber face to face:
"FBI agents quickly came to Bauman’s bedside. A man in sunglasses and black baseball cap had walked right up to him, placed a black backpack on the ground and stepped away, Bauman remembered.
His tip became a critical lead, according to law enforcement officials."
A story that should be about the around the clock work of human investigators and the resolve of a man whose legs were torn off has been turned into a promotional campaign for technology vendors.
Technical Barriers to Facial Recognition
The images captured on the street, both from surveillance cameras and smartphones, were not good enough to deliver high accuracy matching. Recall the video from the Lord & Taylor's surveillance cameras, excerpt below:
There's too much downtilt plus the subjects are tending to look the other way, plus the pixel density is too low (standard definition camera with a ~15 foot wide FoV). And obviously the other suspect (#1) was wearing a forward facing cap and sunglasses, causing even more problems.
Note, this does not mean the cameras deployed were necessarily 'bad'. They are sufficient for general monitoring but not for the radically more demanding needs of facial recognition.
Only 'Hollywood' make believe facial recognition could use this.
Ironically, the images from smartphones were of better suitability for facial recognition than surveillance cameras. Here's an excerpt from one of the best ones:
This is obviously from a much wider FoV and the resulting pixel density is too low to deliver accurate facial matching. A fuzzy image of a face may be 'easy' for a human to verify against a specific person/photo but ask a computer to do this against millions of images and it becomes an exercise in futility.
Other Reports on Video Analytics
Most Recent Industry Reports
The world's leading video surveillance information source, IPVM provides the best reporting, testing and training for 10,000+ members globally. Dedicated to independent and objective information, we uniquely refuse any and all advertisements, sponsorship and consulting from manufacturers.