Real time QoS control for video tracking -

Abstract

In this thesis, we propose a method to match a set of logos in a video in real time without degrading the quality of the video. The matching algorithm should take into account frame rate standard so that we avoid skipping frames which drastically reduces the quality of a video. The contribution of the thesis is two-fold. First, we propose a solution based on fine grain Quality of Service (QoS) control so that to increase predictability of execution times. Our method allows adapting matching logo algorithm by adequately setting quality level parameters for it. The objective of the quality management policy is to meet QoS requirements while maximizing the utilization of available time budget. This allows us to match a set of logos by adapting the quality of the matching algorithm depending on the available deadline. Depending on the progress of the computation (actual time), our method uses a quality manager that chooses the next quality level parameter. Second, we use multithreading in order to parallelize matching of different set of logos, and hence maximize the number of logos that could be matched within the deadline. Our method is fully implemented using C++ and OpenCV library. We present experimental results showing that using our method we can match more logos while respecting the frame rate of a video.