Rozpoznávání lidských činností pomocí Deep Learning

Abstract

The theme of this bachelor's thesis is human activity recognition using Deep Learning. The thesis describes principles of neural networks and methods for action detection and classification. This is followed by implementation, in which two convolutional neural networks based on LeNet and GoogLeNet architectures are created and tested. For testing, human action datasets MSR and UTKinect were used. Implementation was done using programming language Python in combination with the framework Tensorflow.