Classifying human body postures by a two-neuron fuzzy neural network

Human behaviour analysis is an important step of developing a surveillance system. Classifying postures is a attractive topic of human behaviour analysis. Many classifiers and features extracting from human body are developed. However, complex computing is a difficult for implementing models and understanding extracted features. In this paper, we proposed a fuzzy neural network including two neurons to implement easily. Moreover, two features extracted by counting pixels of human body's silhouette are presented. Experiments classify four postures including standing, lying, sitting, and bending. To prove the effectiveness, our model is compared to competing models. Results show the proposed model is better than compared models and improve significantly the accuracy of classifying.