Continuous real-time recognition of basic gestures using hidden Markov models

A system for real-time recognition of 14 basic gestures of the Russian sign language was suggested. The main feature of our approach is the ability to find the beginning and the end of the gestures in a continuous hand movement and to classify them. Segmentation was produced using such characteristics as the velocity and the direction change of hand movement. Hidden Markov models (supervised machine learning approach) were used for training and classification. Besides trajectory, the location of the movement was detected with respect to parts of the body. A software system of gesture recognition with depth sensor was developed for testing our method. The experiments showed successful results in 95% of cases.