This project is from cornell.edu which makes use of skin detection on an Altera D2 development board to track upper body movements. Body movements is captured using a camera in a form of streams – the streams then gets filtered and averaged, after then will be stored on a down-sampled memory block – the down sampled memory block will minimize the space requirement of the original stream (a portion of the stream would be enough to detect motion).
The Altera D2 has a Altera Cyclone® II 2C35 FPGA that uses the down sampled frames to compute for the position of the head, chest and arms. The computed information is then used to construct a 3D robot (ooh yes its a robot!) which is displayed on a VGA screen. The 3D robot is made out of many 3D boxes – head, body and arms. The projection also change depending on the users view of the capturing camera – this is to give the robot a more realistic feel.