The biggest problem facing the implementation of police body cameras is the sheer amount of boring video they produce. Finding a specific incident is possible when the exact time is known, but analyzing the footage to understand how police can work more effectively or prevent altercation has thus far been an impossible task.
Taser International, the military hardware company that essentially owns the police body-worn camera market, believes the solution lies in artificial intelligence. It has acquired a startup called Dextro to build an AI research lab focused on developing tools that make it easier for police to search and analyze the massive video libraries hosted by Taser. Dextro has previously worked with companies like Periscope, the livestreaming service bought by Twitter, to understand what was happening in users’ live videos.
Building artificial intelligence that can understand what’s happening in videos has become a standing area of research, both in large tech companies and academia, but researchers haven’t had access to vast quantities of data from police body cameras. As part of Taser, Dextro will be able to work with 5 petabytes of footage. “Basically nobody has close to even one [petabye],” says David Luan, CEO and cofounder of Dextro.
Dextro has developed a method of analyzing video that identifies the actions taking place in video based on how objects move from one frame to the next. Other popular approaches treat video like still images, looking at frames one at a time and inferring motion. Luan says that approach doesn’t work when trying to understand the actions in videos, rather than just the objects inside. Its research will be specific to law enforcement body-camera footage to create technology that can process police departments’ catalogues as well as video livestreamed from body cameras.
“Usually the actions of interest that happen in body-worn camera footage is unique in that it’s not security footage or classic YouTube-style video,” he said. “The action isn’t guaranteed to be on-screen at all times. It could only be a small portion of a corner. The camera’s perspective is always changing.”