Kinect trained to keep an eye on you
Gesture recognition has taken a step further as researchers from Cornell University develop a program that can “train” a Kinect device to recognise activities and work out what humans are doing. It is conceivable that the solution could be implemented in digital signage to provide inexpensive yet sophisticated diagnostics and allow advertisers to effectively tailor content.
Researchers suggested that such a system could be integrated in smart houses to allow robots to assist with certain tasks.
The researchers took a Kinect device as an input sensor and developed learning algorithms to determine a person’s activities. The algorithm considers an activity to be comprised of a set of sub-activities.
The algorithm was tested to detect and recognise 12 different activities performed by four people in different environments. It achieved an average performance of 84.3% when the person was seen before in the training set and 64.2% when a person was not seen before.