Detekce anomálií v lidském chování

Abstract

The purpose of this thesis is to create an application, that will recognize anomalies in human behavior from video. In theoretical part, it describes how actions can be recognized, different views of the scene in 2D and 3D, depth maps and a description of the functioning of neural networks.

The practical part describes my work, ie the recognition of poses in the video, subsequent prediction and evaluation of behavior prediction. The goal is to recognize when it is an anomaly and when it is normal behavior.