A local neural network approach to detect actions of human body

Local events in a video sequence can be captured by local space-time features, these features helps to extract motion descriptors which capture motion sequence in the video. In the proposed work, the actions of human body is detected by detecting interest points called STIP, in each frame of the video and for each interest point, a motion descriptors called HOG are extracted around each interest point. The dictionary of Bag of Visual Features (BoVF) is created by using HOG descriptors from which normalized histograms are constructed for each action video to train and test the classifier. Feed forward neural Network (FFNN) with Backpropagation classifier is used to classify the actions of human body.