Evaluation of human body tracking system for gesture-based programming of industrial robots

Is low-cost tracking precise enough for recognition of pointing actions? We investigate the quality of the human body tracking available with a Kinect camera by comparing it to a state-of-the-art motion capture system. The application is action recognition with parametric hidden Markov Models (PHMMs) for programming industrial robots. The data from the Kinect is overall more noisy and potentially requires calibration. In conclusion, though, we believe the quality will be sufficient for use with PHMMs.