Gesture recognition for a cooking assistant system

Our objective to develop a system that recognizes cooking gestures to display cooking information in real time. Hence, we construct models for cooking gesture recognition. Our approach extracts joint positions using Kinect and constructs a model using hidden Markov models (HMMs). In this study, we use OpenNI/NiTE to extract the features, and the Hidden Markov Models Toolkit (HTK) to construct HMMs. When we select the torso and elbows as the combination of joint positions, the structured models have a 90.6% gesture recognition rate.