Algorithms and methods of motion tracking in video flows

Introduction

In our time, there is an ever-increasing amount of information generated. However, with the increasing volumes of information, it becomes more difficult to work with it, in addition, there are new types of information, efficient processing algorithms, which do not exist yet. These types of information include video streams. Till nowadays, the number of businesses companies, witch are using video streams in their work was not very big, as well as the volume of this information, but there is a trend of increase in the volume of video, and to increase the range of its usage. Incensement of video information reflecting real events in the world also increases time required for processing and analysis. In addition, quite often, video contains events which analysis needs highly paid professionals capable of accurately and efficiently describe such events. That raises the question: "How can we reduce the time and material costs, in order to analyze video from a variety of sources, such as television, Internet, or home videos?" To resolve this problem, it is necessary to provide video as a graphic database, and build searching algorithm events allowing computers to automatically detect and analyze events contained in the stream. Note that the implementation of such a change may be difficult, if an analysis should be conducted in real time, for example, simultaneously with the images broadcast on television. In this case, the time allotted for the search algorithm events should be limited, and that consequently, increased demands on the performance of computers.

Relevance

This work addresses task of the development of existing ideas and finding a solution to the detection of events in video stream and recognition tasks in applications. This paper will further develop this algorithm: tracking events in video stream by current and basic frames.

Purpose and objectives

Objective - examining available methods to achieve recognition events in video stream point out their shortcomings and propose ways to improve them. The work has to increase speed of the method, which is based on the deduction of frames, with a possible increase of the required RAM.

Alleged scientific novelty and practical value

Developing system “Video football match” must be able to keep track of the players during the match, as well as provide statistics for each of the players. The improvement will be subjected to the method of allocating players figures - each frame will be split into two, containing, players presenting different teams, which will reduce the number of conflict situations accelerate the program.

National research and development

The task of detecting moving objects in real-time has a long history, but, because of its nature, it still lacks a clear decision. First of all, the detection conditions may be different. For example, if you have to deal with the black and white binary image, the task is easier compared to the case where the input stream contains color raster with smooth color transitions from one hue to another. The image may come from statically located cameras and all streams will have approximately the same background with possible differences in lighting. On the other hand, camera, located on a moving object can withdraw another stationary object to it. Noise levels can vary significantly. Natural phenomena, such as rain, snow, fog, wind, etc. could play a significant element of volatility initially stationary scene. All this makes algorithms, excellent working in some circumstances, entirely unfit for others. Algorithms of detection and analysis usually require stability in a wide range of very different external conditions. In general requirements for such algorithms as follow. 1) Low computing complexity and work in real time. 2) Sustainable detection at different times of the day with artificial lighting. 3) Stable work at any time of the year in all weather conditions. A detailed discussion of algorithms detect moving objects: - frames difference; - basic frame method; - methods of mathematical morphology; - methods of allocation; - correlation of images; - methods rail.

Development and research, planned to perform

To explore the possibilities of different algorithms of recognition events in video frames, valuate the effectiveness of different approaches, and choose the best from the ratios speed-quality;

• use division a frame on two;
• examine the

results of segmentation and figure whether it is efficient or not.

On this stage author has the following results:

Summary and Conclusions

As a result of this work has been given to basic techniques and algorithms of recognition events in video stream and their main drawbacks are shown. It also considered a way of improving the work of identification algorithms by increasing the required RAM.

Literature :

1. N.S. Baygarova, Y.A. Buxstab, N.N. Evteeva, Modern technology search in the collections of electronic images. / / Electronic Library 2001, Issue 4, Volume 4.
2. Jain, R. and Gupta, A., Visual Information Retrieval, Communications of the ACM, 1997, vol. 40, no. 5.
3. N.C. Baygarova, Y.A. Buxstab Some principles of the organization seeking video. Programming, No. 3, 1999, pp 165-170