Facial Recognition Failed to Find Boston Bombers21 Apr 2013
During the now-infamous hunt for the Boston Marathon Tragedy suspects, the Boston Police Commissioner stated that the technology provided to them by some of the biggest vendors in the world failed to recognize the bombers – even with their faces having been stored in all of the official databases necessary for the software to work.
Facial Recognition Software companies make millions selling their software to state and federal agencies in the promise that in such an event as this, their tools would be able to find and recognize even the worst of offenders with the most grainy of photos.
What we now know is that all of the promises made over the years that this technology is sound and works fell short and the case was finally solved by people looking at hours of footage and witnesses being shown a number of potential images and even public involvement.
What does that mean for the future of this kind of software? It may be that all of the facial recognition demos that we have seen in the news and demonstrated in movies have come from false assumptions that this kind of technology is easy to make. Even Facebook, which actually has some of the best facial recognition technology in the world, sometimes gets it wrong when it asks you to tag someone in a photo.
The question that needs to be asked is: what makes it work and what doesn’t? As a software engineer, I can tell you that more data is the key, but not the kind of data that the government retains in their large personnel databases. The kind of data needed is also the kind that is actively tracking and identifying people not just in a mug shot but in walking down the street or into a store.
Collecting this kind of data that immediately identifies people as they are moving would be considered a tremendous invasion of privacy, so it’s doubtful that these expensive vendors will get any better anytime soon without someone in a higher government position making some serious potential constitutional violations. It looks like, for the time being, we will simply need to do it the good old fashioned way: using the human brain. And this time, it thankfully worked.