Metody detekce pohybu ve videosekvencích

Abstract

This thesis describes basics of background modeling of video sequences and its further foreground segmentation. Several mostly used methods in this part of computer vision are shortly described. Next part describes in detail basic recursive multimodal version of mixture of gaussians method. Mentioned are its common problems when used in real life and their possible solution. Part of this work is implementation of this method in C language with usage of OpenCV library and also an implementation in open computing environment OpenCL. Process of both implementations of this algorithm is described in detail, including all necessary changes for implementation in OpenCL. Both variants were tested for performance on two video sequences and the resulsts are shortly discussed. The final part then evaluates the mixture of gaussians method and proposes some additions for the implementation.