Human action recognition based on neural network

In this paper based on neural network human actions are recognized. The action video is based on learning spacial related human body posture prototype using self organizing map. Self Organizing maps(SOM) is used to learn the human postures. SOM can be constructed based on three procedures they are competition, cooperation, adaption. In feature extraction the action recognition is expensive. So we use multilayer perceptron for action classification. fuzzy distance can produce a time inavariant action representation. Action are classified based on multilayer Perceptron. Inorder to recognize the action multi cameras are used. Action recognition is performed for each of N cameras by using MLP ie, feed forward neural network. Back propagation algorithm is trained in MLP. These actions are viewed from different angles, a view invariant action is recognized. The proposed method can be applied to videos depicting interaction between humans without any modification.