It's an old movie trope: the pursued loses their pursuer in some
anonymous crowd, usually a street market or semi-ambiguous ethnic
festival. Relieved, they fade into the sea of bodies and faces as a wave
might collapse into the sea. The pursuer steps into the throng, gun no
longer at the ready, and gives a perfunctory glance around before making
that humiliating phone call to headquarters: I lost him.
It's cool, bad guy. Computers have trouble with crowds too. Individuals blend together when they're surrounded by a great many other individuals, as the reality that we have more in common with other humans than we don't takes hold. As it stands now, if some surveillance software identified a certain individual, whether they're hustling through a train station with a briefcase bomb or giving riot cops the slip, it would be very difficult to continue tracking them among other individuals. Typically, after only a few seconds of identification, current software will lose its subject(s).
This was the problem facing researchers at the Spanish National Research Council (CSIC), as detailed in a new study published in the journal Nature Methods describing a piece of software (already patented) that would appear to have overcome the challenge. Using a set of newly discovered algorithms, the team was able to successfully sort through different members of an observed group, in effect identifying each one of them individually.
While surveillance is the most obvious application (to most/many of us), the researchers are more interested in the newfound ability to examine and categorize the interactions of social groups among different animal species. This is what the patent for their tool, called idTracker, actually covers. We know that many animals make decisions based on the group behavior of their respective communities, but what the rules of those interactions actually are remain difficult to discern. Being able to "tag" individual members with this software means being able to acquire large amounts of data on how those members interact and, eventually, what might predict those interactions (the rules, in other words).
A popular sentiment these days can be summed up as such: less algorithms, more human intelligence. Algorithms have become extremely public in a short amount of time—without much corresponding education on the subject—and we can see them making decisions all around us, from the movies we might like to watch to whether our banking activity is normal or suspicious. What gets lost in this notion is that algorithms, such that might be implemented in surveillance software, aren't just replacing humans, but are doing tasks that humans are simply ill-equipped to do themselves. Crowd-sorting is one such example.
The reason is easy to see: tracking an individual in a crowd is not simply a matter of identifying that one individual. It means identifying every individual or at least some very large portion of those individuals. The software works by taking note of crowd/community members that briefly branch away into relative solitude. Once the member is away, a digital "footprint" is created composed of their distinct features. If you're able to collect enough features from enough individuals, the crowd will instead become a shifting map of footprints. The human visual system is not up to this task.
"In the short term, this will be used in science," noted co-author Alfonso Pérez Escudero in a statement, "but in the longer term, the method we have developed can be applied to recognize people in large crowds, vehicles or parts in a factory, for instance."
It's cool, bad guy. Computers have trouble with crowds too. Individuals blend together when they're surrounded by a great many other individuals, as the reality that we have more in common with other humans than we don't takes hold. As it stands now, if some surveillance software identified a certain individual, whether they're hustling through a train station with a briefcase bomb or giving riot cops the slip, it would be very difficult to continue tracking them among other individuals. Typically, after only a few seconds of identification, current software will lose its subject(s).
This was the problem facing researchers at the Spanish National Research Council (CSIC), as detailed in a new study published in the journal Nature Methods describing a piece of software (already patented) that would appear to have overcome the challenge. Using a set of newly discovered algorithms, the team was able to successfully sort through different members of an observed group, in effect identifying each one of them individually.
While surveillance is the most obvious application (to most/many of us), the researchers are more interested in the newfound ability to examine and categorize the interactions of social groups among different animal species. This is what the patent for their tool, called idTracker, actually covers. We know that many animals make decisions based on the group behavior of their respective communities, but what the rules of those interactions actually are remain difficult to discern. Being able to "tag" individual members with this software means being able to acquire large amounts of data on how those members interact and, eventually, what might predict those interactions (the rules, in other words).
"From now on, we will be able to quantitatively determine the rules of animal behavior in groups taking into account the individuality of each animal," said Gonzalo G. de Polavieja, the study's lead researcher.
A popular sentiment these days can be summed up as such: less algorithms, more human intelligence. Algorithms have become extremely public in a short amount of time—without much corresponding education on the subject—and we can see them making decisions all around us, from the movies we might like to watch to whether our banking activity is normal or suspicious. What gets lost in this notion is that algorithms, such that might be implemented in surveillance software, aren't just replacing humans, but are doing tasks that humans are simply ill-equipped to do themselves. Crowd-sorting is one such example.
The reason is easy to see: tracking an individual in a crowd is not simply a matter of identifying that one individual. It means identifying every individual or at least some very large portion of those individuals. The software works by taking note of crowd/community members that briefly branch away into relative solitude. Once the member is away, a digital "footprint" is created composed of their distinct features. If you're able to collect enough features from enough individuals, the crowd will instead become a shifting map of footprints. The human visual system is not up to this task.
"In the short term, this will be used in science," noted co-author Alfonso Pérez Escudero in a statement, "but in the longer term, the method we have developed can be applied to recognize people in large crowds, vehicles or parts in a factory, for instance."
No comments:
Post a Comment