德国计算机硕士论文：a user interface for efficient entertainment

Abstract:摘要

本文提出了一种简化视频摘要的用户界面。本文重点研究了一个简单背景下的连续视频目标跟踪问题。采用基于颜色直方图的目标跟踪方法。从直方图中提取颜色信息，通过余弦相似性度量计算图像之间的相似性。提出了一种基于目标帧最后位置的搜索方法。我们已经评估了四种不同类型的视频中的用户界面，演示结果验证了我们的用户界面在连续视频中准确跟踪目标的能力。In this thesis, we presented a user interface to ease video summarization. The thesis focuses to track a target with simple background from a consecutive video. A color histogram based target tracking method is applied. The color information is extracted from the color histograms for computing the similarity between images through cosine similarity measurement.  We proposed a simple approach to search the target, which is based on the target’s position of last frame. We have evaluated the user interface in four different kinds of videos, the demonstrated results validate the ability of our user interface to track the target accurately in consecutive videos.

介绍：
相关工作：介绍
用户界面：
框架重要功能：
目标跟踪：
算法改进：
评价：

Introduction：
Related Work: introduction
User Interface:
Frame Importance Function:
Target Tracking:
Algorithm Improvement:
Evaluation:

Conclusions:


In this thesis, we have presented an effective user interface software for frame weighting  and  target  tracking. 
The frame importance function is designed based on the ffmpeg library. It can completely meet the users’ requirements. The weighting curve will help the users to select and save the interesting section of a video intuitively. 
Target tracking is the main focus of this thesis, we develop an efficient algorithm according to the strategies of “color histogram” and “cosine similarity”.  The advantages of our algorithm are simple, fast and easy to implement, as well as low cost of computational expense. We successfully explored the matching area in consecutive frames, and handled with the case of mutational position. Comparing with other color histogram based algorithms, our method expanded the vector length (dimension) creatively, and the accuracy is also improved successfully.
One limitation of our software is the function of video segmentation only suits one section of a video. The users can only segment the video iteratively if there are several interesting sections. We believe that this problem could be solved by conducting more deeply fundamental researches and technical explorations.
Another limitation is caused by the searching method of the target tracking algorithm. For the moment, our searching method can only weight a small and fixed value for the distance between centers. The effect of the small weight is obviously limited. In some cases, the large weights could lead to errors. In other words, the software will select areas with minimum distance rather than maximum similarity value.