Rozpoznávání lidských činností pomocí Deep Learningu

Abstract

This bachelor thesis deals with convolutional neural networks and their usage in terms of human action recognition. The first part covers fundametals of convolutional neural network and deep learning. In the next part topics of neural network from the human action recognition point of view are described aswell. This knowledge is furthermore used for an experiment where ResNet network is used for training on both MSR and KARD datasets alike. The next part of the experiment covers a method for data generation on said datasets and its comparison with default training. In the experiment's conclusion the benefit and possible drawbacks of this method are evaluated.