Detekce anomálií v chování řídiče

Abstract

This master thesis describes anomaly detection using the OpenPose tool and neural networks. The main goal was to implement an autoencoder type neural network and to perform experiments with the architecture of this network. The beginning of the document is focused on the prerequisites. Then, the used neural network architectures and their training on the acquired data are described. Finally, the testing of the success of the neural networks is discussed. As a result, this thesis compares the success of several architectures in detecting anomalies.