Detekce anomálií

Abstract

The topic of this master thesis is detection of anomalies in video sequences using machine learning. It primarily focuses on the analysis of the vehicle driver's condition. The introduction outlines the problematics and application of such a system. Following theoretical part deals with the phases of the anomaly detector and possible approaches to their solution. It is divided into three main parts, which are detection of the object of interest, data representation and data analysis. Following is a list of publications related to this topic. Practical part is focused on own implementation of the vehicle driver anomaly detector. It includes description of used dataset, choice of approach in each part, summary of the proposed method and the results achieved. At the end of the work, the possible use of the application, its shortcomings and possibilities for further development are described.