Recognition of human walking/running actions based on neural network

High precision recognition of human actions directly from video records is still open problem. In this paper an approach for human action recognition for full body analysis based on a novel configuration of convolutional neural network is presented. The proposed convolutional neural network approach is a variant of multilayer perceptron, which main advantage is its ability to learn the feature extraction layers during retropropagation of errors from the lower layers using as input an image without any pre-processing. It permits to introduce the human descriptive action extraction process directly to neural network for more fast recognition. In order to evaluate the proposed approach a framework for recognizing human walking/running actions has been designed and tested on developed dataset that consists of multiple video records providing 4000 images per activity used for motion detection and activity interpretation.