Rather than make use of the depth- and color-images from the Kinect device, the researchers chose instead to make use of the high quality skeletal model which is output from the sensor.
This skeletal information was first recorded, a set of features extracted from it and a collection of machine learning algorithms called Weka used to help identify a person on the basis of the previously recorded training data.
Using the Kinect-based system, the researchers not evaluated a number of body features together with step length and speed to determine how effective they were for person identification. They found out that four features -- height, length of legs, length of torso, and length of the left upper arm -- were sufficient to correctly identify a person in over 90 per cent of all cases.
A technical paper -- Gait recognition with Kinect -- which describes the research was presented at the First Workshop on Kinect in Pervasive Computing which was held in Newcastle, UK on June 18, 2012. It can be downloaded here.
-- Dave Wilson, Senior Editor, Vision Systems Design