(Adds detail about surveillance cameras in eighth paragraph)
By Sarah McBride
April 19 (Reuters) - As investigators searched through videos and photos for clues that ultimately led them to the suspects in this week's bombing at the Boston Marathon, their search likely relied more on old-fashioned police work than technology, officials who work on such investigations say.
Advances in facial-recognition technology and related fields have been significant in recent years, but the technology mostly still isn't good enough to match blurry images or those that show a face only from an angle, say people familiar with law-enforcement dealings.
"They probably just applied a lot of eyeballs to this problem," says Todd Waits, digital investigation and intelligence director at Carnegie Mellon University's Software Engineering Institute and a specialist in forensic video techniques. "It would be a very manual process at first."
Investigators would have likely started simply by looking at images of the blast itself and worked backward, looking for any type of suspicious activity. On Thursday, FBI Special Agent Richard DesLauriers said that law enforcement had observed images of a suspect setting down his backpack at the site of the second explosion.
One suspect Tamerlan Tsarnaev, was killed in a shoot-out overnight. His brother, Dzhokhar Tsarnaev, was captured in a Boston suburb Friday evening, police said.
Based on similar investigations, technology likely will take on an increasingly important role for the next stages, particularly when it comes to examining the suspects' computers and mobile phones, forensics experts say.
"We would analyze computers, find evidence of Internet searches for bombmaking, that kind of thing," said Detective Andrew Kleinick of the Los Angeles Police Department's computer crimes unit. "Look for emails to see if they've been corresponding with anyone who might have radicalized them."
In the future, facial recognition technology is likely to become increasingly relevant to such investigations, enabling law enforcement to match imperfect images of people captured by video or cameras to databases of mug shots or passport photos. Higher resolutions in newer surveillance cameras should help too, particularly if they are matched by more storage capabilities in those cameras, Waits said.
Over the four years from 2006 to 2010 accuracy in facial-recognition algorithms has improved by a factor of 10, according to the National Institute of Standards and Technology.
Some of the biggest improvements lie in algorithms that correct for lighting, pose, and expression, says Brian Martin, director of biometric research at the Safran Group's MorphoTrust, an identity-services company that works with law-enforcement agencies.
The advent of unmanned drones with high-resolution video cameras, deployed over high-traffic events, could also give law enforcement important new tools.
"Once they become more commonplace, you'll be able to honestly go back in time and say, OK, this vehicle or this person came from over here, went there, entered this subway stop," says Chris Westphal, chief executive of software company Visual Analytics. "You'll be able to replay."
While drones have proven controversial because of privacy concerns - Charlottesville, Virginia, recently banned their use - they are likely to prove popular with law enforcement because of their low costs and efficiency.
Expect them over key sporting events, parades and the like in around five years, Westphal says. (Reporting by Sarah McBride in San Francisco; Editing by Jonathan Weber and Lisa Shumaker)