Skip to Main Content
This paper presents performance of view-based approach in automated recognition of predefined hand and gross body actions using artificial neural network. This approach represents motion by a static grey scale image template computed by collapsing the temporal components into the cumulative image-difference of frames. The seven invariant Hu moments are used as the feature vectors. The performance of the system is tested in real time using feed forward multilayer perceptron (MLP) based on back propagation.