Our objective is to implement human action recognition in video streams through
learning models. We developed a 3D convolution neural network model which
can learn spatio-temporal features and classify human actions without any prior knowledge. We extended 2-D CNN for image recognition to 3-D CNN for video recognition.Our model can be easily modified in terms of number of nodes in each layer.Our 3-D CNN model converged after 18 epochs.We tested our 3-D CNN model on Weizmann dataset
containing ten classifications and our model gives comparable accuracy with recent research in this
field
|