Detekce anomálií v lidském chování

Abstract

Detection of anomalies in human activities is an important task with a wide range of applications. This thesis is focused on the detection of anomalies in human activities using neural networks and skeleton features. Detections are performed on drivers of motor vehicles. Concerning this type of task, it is usually very difficult to obtain a dataset in which individual anomalies can be recognized. Thus, the main goal is not to distinguish individual anomalies from each other, but to determine whether there are anomalies at all, and to detect them as precisely as possible. In the theoretical part of this thesis, the methods of human detection are discussed alongside the principles of neural networks and some key technologies for anomaly detection. The practical part describes the design and implementation of several detectors in the Tensorflow environment, including experimentation with various skeleton features and subsequent evaluation of the results and accuracy of the created system.