Gesture recognition system

Abstract

The goal of this research is to improve the algorithms for gesture recognition based on the analysis of a two-dimensional space model using neural network technologies. Completed analytical review of existing software solutions and approaches to the implementation the gesture recognition systems, including detection, tracking and recognition stage. Also block diagrams of the gesture recognition module were drawn up, appropriate descriptors and classifers have been selected for better results. To reduce the dimensionality of the feature vector, it was decided to employ a special associative network with a bottleneck and multilayer perceptron with a probabilistic principal component analysis. Also evaluated the efficiency of gesture recognition module.