Analysis Algorithm Of Gesture Recognition Based Dynamic Fuzzy Neural Network

Gesture is a natural intuitive and interactive tools, which plays an important role in high demanding human-computer interaction. It is not only a communication channel, but also a demonstration study carrier. By using gestures, people can naturally communicate with robot, some difficult and high dangerous mission can be completed by robot according to people's intentions, which can reduce the risk, and people think that it is very interesting and value. But now using gestures of human-computer interaction still not completely, and is of high useful value, so recently gesture recognition becomes a research hot spot and difficulty spot.From the Angle of man-machine natural interaction, the article studies the independent of user perspective of gesture recognition method, and proposes a gesture recognition algorithm based on dynamic fuzzy neural networks, this method is mainly use for the static gestures. Considering the positive gestures do not fully describe natural characteristics, this paper uses the perfect gesture set to describe the feature set, according to judge hamming distance of Zernike moments gesture features of the original image and Zernike moments gesture features of reconstruction images define the highest order of Zernike moments. In dimension reduction of gesture feature, from the angle of classification accuracy and maintaining the geometric properties among the original image point, this paper uses Isometric mappings (Isomap) method for gestures features dimension reduction. In this paper, classifier based on dynamic fuzzy neural network is constructed, used to classify input samples. In recognition phase, test set features consider training samples characteristics as input data of classifier.Next, the paper analyzes the performance of algorithm of gesture recognition based on dynamic fuzzy neural network and proposes the existing problems and solutions. Finally, through simulation experiments, this paper proof that perfect gestures set based on Zernike moments can completely represent gestures characteristics and Isomap technology was also proved it can save geometric properties between original image points. Experiments show that the performance of classifier based on dynamic fuzzy neural network is better than RBF classifier, Hopfield classifier, BP classifier and LQV classifier. The final result of experiment shows that the proposed algorithm recognition rate is good in this paper.