Systém pro čítání osob ve videosekvencích

Abstract

The goal of this thesis is to create a system, that can determine the number of people, that are present in the input images. The thesis starts by describing both historical and state of the art solutions used to resolve this problem and datasets that are used in this area. Based on this information a neural network was created. It`s goal is to create a density map, which describes the density of the people captured in the input image. Based on this density, the number of people is estimated. The neural network creates this map from a consecutive sequence of images. The thesis is further concerned with the question, whether the use of this sequence has any benefits as opposed to using a single image. At the end, the results are evaluated and further approach and other possible modifications to the neural network are discussed.