Sensor To Monitor TV Watchers Demoed At Cable Labs 302
An anonymous reader writes "Cable operators at the semi-annual CableLab's Innovation Showcase have informally voted as best new product a gizmo that can determine how many people are watching a TV. Developed by Israeli company PrimeSense, the product lets digital devices see a 3-D view of the world (the images look like something from thermal imaging). In other words, that cable set-top box will know whether three people are sitting on the sofa watching TV and how many are adults vs. children. Do we really need cable and/or video service operators knowing this? It all happens via a chip that resides in a camera that plugs into the set-top box."
Re:I bet it doesn't work! (Score:5, Informative)
On the contrary. I bet it does.
One of my graduate professors at the University of Missouri is doing some work with elder care with technology like this. Getting real 3-D information from multiple cameras takes a lot of processing time, but they can segment the space in a room down to 1 inch cubes. The result is a 3-D silhouette of the objects and people moving in the room. They can tell the difference between people moving throughout the room. A small child is different from a dog in that the dog has 4 legs, for example.
They can identify whether a person is laying on the couch or has fallen. They can extract information such as the bend of the spine and whether a person favors one leg or the other. A silhouette of a cardboard cutout would appear flat to the camera.
What I described above is PHD research using some fairly complex computer vision, 3-D segmentation and pattern recognition algorithms. 3-D scene reconstruction cannot be done with a single camera. The math doesn't work. I would not expect a set-top box with a single webcam to be that good. I would, however, expect them to do motion segmentation on frames of video. Background subtraction would let them ignore the furniture in the room and identify regions of motion. From there, pattern recognition algorithms could be used to find faces and identify the relative shape of a body based on a template. Given that, you could identify whether a person was tall or short and the relative proportions of their bodies. You probably couldn't identify male or female, though. That would be a tough call. You probably could identify a dog versus a child with relative ease.
Luckily, the amount of bandwidth needed to send these images back to the cable company would be pretty massive (given everyone who watches cable in a small to medium sized city) so you shouldn't worry about that. You probably only need to worry about packets being sent back over the cable line identifying the date, time, number of adults, number of children, channel and number of seconds since the last channel change.
Re:duct tape (Score:3, Informative)