Rozpoznávanie ľudských činností z videa pomocou neurónových sietí

Abstract

This bachelor thesis focuses on the recognition of human activities from video recordings using neural networks. The aim of the thesis is to design and implement a system capable of recognizing selected human activities based on the movement of a person in video data. To obtain information about human movement, the MediaPipe library is used, which enables the detection of key points of the human pose. Based on the data, temporal sequences are created and subsequently processed using an LSTM neural network. The proposed system was tested on a set of videos containing different types of activities and achieved reasonable recognition accuracy, with the best model reaching approximately 89 % validation accuracy.