Static Gesture Recognition Based On Hidden Markov Model

Gesture recognition is one of the key technologies in the field of human-computer interaction and computer vision, it is widely used in a variety of domains such as the Motion Simulation Games, Mechanical Control, Augmented Reality, Smart Homes, Auxiliary System, Sign Language Communication, Virtual Reality, etc. Unlike the traditional way of human-computer interaction, human-computer interaction based on the technology of gesture recognition shows us a more instinct and effective way to get on well with virtual environment, as well as satisfies people with higher demand for intelligent level. Therefore, there is important and far-reaching research significance of simple and efficient gesture recognition technology.This paper puts forward a new static gesture recognition algorithm based on hidden markov models. Which adopts two kinds of new features-specific Angle profile entropy feature and upper contour feature, and then they are respectively used for training of hidden markov model parameters, finally identify gesture categories hierarchically. In the course of feature selection, considering both classifier and its research status synthetically, we extract not only specific Angle profile entropy feature representing local gesture information and upper contour feature representing global gesture information,but also two old features-specific Angle area features and contour feature to make contrast experiment to test the effectiveness of the new features. In the course of classification, in order to use Hidden Markov Model which is simple, easy to expand and has the advantages of time scale invariance when dealing with dynamic sequence, this algorithm simulates the static characteristic into dynamic sequence. Since Hidden Markov Model is suitable for processing of a sequence, so after reconfiguration, Hidden Markov Model, trained by static gesture feature, has space scale invariance. Our experiment shows that:this algorithm works well even for gestures whose shape difference is not big and provides a independent training for each gesture category which is good for system expansion; Two new features complement each other and recognize gestures step by step, which can reduce the feature dimension in a way, make gestures to get a complete description of the shape characteristics. However, it is not ideal enough when it comes to some special gestures who are easily mixed in shape.Focusing on those unsatisfied result, in order to further improve the recognition effect, this paper adds texture energy feature which can reflect the internal details of the gesture image and make his final correction estimation based on minimum total error probability. The experimental results show that, after further revision, the method has good recognition effect for gestures whose shape differences are small as well as has strong real-time performance.