Forensic video analysis remains a time-consuming, human task.
As investigators try to figure out what happened today during the bombings at the Boston Marathon, they'll turn to video taken at the scene of the explosions.
In addition to any closed-circuit television cameras lining Boylston Street and its surroundings, The Bureau Chief of Public Information, Cheryl Fiandaca,called for members of the public to send in video from near the finish line.
Once the police have the prospective evidence in hand, they'll need to run forensic analysis on it. What we know from recent years is that the amount of evidence can be staggering large. For example, the Vancouver riot in early 2011 brought 1,600 hours of video streaming into that city's police department. Just to watch all that video through one time is a substantial task, let alone examining it closely or trying to find events or people of interest.
Right now, there is no video software that can do this type of analysis, not even in a first-pass way. IARPA (DARPA for the intelligence services) put out a call for proposals in 2010 for this kind of "Automated Low-level Analysis and Description of Diverse Intelligence Video." It described, in brief, the problem that investigators (or intelligence analysts) face:
Massive numbers of video clips are generated daily on many types of consumer electronics and uploaded to the internet. In contrast to videos that are produced for broadcast or from planned surveillance, the "unconstrained" video clips produced by anyone who has a digital camera present a significant challenge for manual as well as automated analysis.
So, at the moment, human investigators must watch and code each and every second of the video that they collect.
NEXT STORY How Kaggle Is Changing How We Work