It seems like every gadget freak, geek, and maybe even Gleek wants to get their hands on Microsoft's new Kinect gaming device, which captures 3-D movement using a camera, monochrome sensors, and infra-red light. But they don't necessarily want to hook the gizmo up to an Xbox and play a game. They want to hack it.
Alexandre Alahi, a Ph.D. student in the Signal Processing Lab at the Ecole Polytechnique Fédérale de Lausanne, was no different. He and a team used open-source software released online just days after the Kinect's launch to develop a new, patented algorithm that leverages multiple Kinects instead of just one to detect crowds—even in the dark.
"I was fascinated by the performance of the camera in assessing the depth of a scene at such an affordable price," says Alahi, a video-surveillance technology specialist.
The algorithm works to combine the viewing angles from multiple Kinects to recognize shapes and differentiate, for example, between a human being, a bicycle, or a vehicle. And whereas the normal set-up only detects up to a few meters, the algorithm expands this scope to tens of meters and allows high-level function in low light without confusing shadows for human figures.
"Even if shapes are superimposed, our algorithm is robust enough to distinguish them," adds Alahi.
View a video about real-time detection and tracking of subjects using multiple Kinect cameras.
Alahi envisions numerous applications for his system, including security at airports and train stations, where it could provide precise statistical information to help optimize the flow of people or be used to detect suspicious behavior. He can also imagine its use to track numerous players on a sports field, number of people in a queue, or customer behavior inside shops, where it could possibly predict behavior.