Squat movement recognition using hidden Markov models

This paper proposes a novel squat movement recognition system using hidden Markov models (HMMs), whose data is captured by Perception Neuron (PN) [1]. Here, the PN generates entire skeleton data of human body in each frame and uses them to form a data stream in real-time. In our experiments, we collect the data streams of 6 squats including a standard one with 5 non-standard ones. Based on the HMMs learned by these data streams, our proposed recognition system can correctly distinguish the patterns of the non-standard squats from the standard one. Our experimental results show that the recognition accuracies of the standard squat and 5 non-standard ones reach high level, respectively.