[Biography] [Abstract]

Introduction

CCTV is a common form of technical security systems. Until about the mid-late 90-ies of XX century the concept of "video surveillance system" meant a hardware system consisting of cameras, equipment, video recording, and terminals with simple controls for viewing video recordings and real- time, and line for communication between these elements. Such systems are called DVRs. Most often, the DVR uses the fixed camera, rarely turning camera. This type of system has two main disadvantages. Firstly, the operators, who are responsible to monitor what is happening in the picture is overlaid by high load, and in practice they can survive only for 2-3 hours, even if it is used only one surveillance camera. Second, the search for events in the video archive, if the time of the event is not exactly known, can be performed only by viewing the whole archive. DVRs are often supplied with the algorithm for determining motion in the frame, which improves using of video surveillance systems. However, a simple image detector is useless in heavy traffic, such as highways or busy streets. Many recent developments in the field of digital image processing can be directed to the problems associated with the use of DVRs.

The problem urgency

Security systems designed to prevent misconduct, are complex facilities and administrative activities. Modern technical means intended to protect, enhance such important figures as timely reaction to the violation, the amount involved in the complex security personnel, the complexity of decision making in the performance and actions required in response to violations. Nowadays, the problem of effective policing has become particularly topical in connection with the increased activity of terrorist organizations, so the task of improving security devices is a challenge.

Purpose and objectives:

The aim - to explore available methods for solving the problem of recognition events in the video stream to point out their shortcomings and suggest ways to address them. The task of the work - to create a software product which effectively detect and parameterize objects frame of video.

Scientific novelty:

Develop algorithms to improve noise immunity and reliability of response and enhance other capabilities of surveillance systems. Existing software systems show low rates of recognition on the video with the increased noise and distortion.

Expected practical results:

After completing the system development will be obtained working program designed to effectively improve the image quality of video for later raspoznavniya. Will be improved methods for preprocessing, image correction, noise removal, increasing the brightness of certain areas.

Research and development:

Problem of detecting moving objects was set for a long time, but because of its specificity, it still has no unique solution. First, the detection conditions may be different. For example, if there is a binary black and white image the problem is considerably simplified in comparison with the case where the input comes full color screens with smooth transitions from one color hue to another. The image may come from the static camera positions, so that all image will have approximately the same background, with possible differences in the coverage. On the other hand, the camera on a moving object, can shoot the other fixed object. Noise levels may vary substantially. Natural phenomena such as rain, snow, fog, wind, etc. can make a significant element of the oscillations in an initially motionless scene. All this makes the algorithms work well in some circumstances, entirely unsuitable for others. Detection algorithms and motion analysis usually are required stability in a broad range of significantly different environmental conditions. In general, the requirements for such algorithms are as follows:

Low computational complexity and work in real time.

Sustainable detection at different times and artificial lighting.

Stable operation at any time of year in all weather conditions.

The diagram shows the basic procedures and methods for processing the initial phase of the perception of the visual field by means of sensors, such as cameras to the final, which is the recognition.

Preprocessing operation used almost always after the removal of information from the video sensor and aims to reduce noise in the image resulting from the sampling and quantization, as well as the suppression of external noise.Typically, this is the operation of averaging and smoothing of histograms.

Segmentation

Segmentation is generally understood as the search for homogeneous regions in an image. This stage is very difficult and in general do not algorithmic through for arbitrary images. The most common methods of segmentation based on the definition of uniform brightness (color) or a homogeneous type of textures.

Methods of building areas are effective if there is a stable connection within the individual segments. Method of separating boundaries are used if the boundaries sufficiently clear and stable. The above methods serve to isolate segments of the criterion of uniform brightness. Also there is an effective method for the watershed based on finding local minimum, followed by grouping them around the areas of connectivity.All methods are quite acceptable in terms of computational cost, however, for each of them characterized by ambiguity marking points in real situations because of the need to apply heuristics (choice of thresholds match the brightness, the choice of digital masks, etc.). Another method is multi-valued partitioning, based on a combination of different techniques to reduce uncertainty. Practical importance have parallel processing algorithms to accelerate the process of marking on the basis of logical analysis of adjacent elements.

To describe the segmentation and image properties, namely uniformity, roughness, regularity, apply texture methods conventionally divide into two categories: statistical and structural. An example of a statistical approach is the use of matrix matches that are generated from source images, followed by calculation of statistical moments and entropy. A structural approach, eg based on the Voronoi mosaic, are constructed as a set of polygons. Polygons with the general properties are combined to the field. To investigate the general properties of the frequently used features - the moments of polygons. After the segmentation arise interferences in the form of separate changes of isolated elements of the image, as well as a distortion of some connected fields. In practice the most widely is used digital filter masks and non-linear filters such as median to remove noise. In the case of segmentation by separating boundaries using averaging filter masks is impossible because the border is not highlighted, but blurred. For underlining the contours is used special operators of integral type.

Recognition

Recognition - most often the final stage of processing, the underlying processes of interpretation and understanding. Input for image recognition are highlighted as a result of segmentation, and partially restored. They differ from the standard geometrical and brightness distortions.

Conclusion

The process of image recognition is a complex multi-step procedure. Multistage (hierarchical) is due to the fact that different processing tasks actually closely related and quality of the solution to one of them affects the choice of method for solving the rest. So the choice of method depends on the recognition of specific conditions presenting the input images, including the nature of the background of other images, noise conditions and is associated with the choice of methods for preprocessing, segmentation and filtering.

Literature

1. Гиренко А.В., Ляшенко В.В., Машталир В.П., Путятин Е.П. Методы корреляционного обнаружения объектов // Харьков: АО “БизнесИнформ”, 1996. 112 с.

2. Chen C.H., Rau L.F., Wang P.S.P. Handbook of pattern recognition and computer vision // Singapore-New Jersey-London-Hong Kong: World Scientific Publishing Co. Pte. Ltd., 1995. 984 p.

3. Путятин Е.П., Аверин С.И. Обработка изображений в робототехнике // М: Машиностроение, 1990. 320 с.

4. Ватолин Д., Обухов А., Гришин С. Фильтр для удаления "блочности" на видео данных / VirtualDub MSU Smart Deblocking Filter [Электронный ресурс] // Ватолин Д., Обухов А., Гришин С. - режим доступа: http://compression.ru/video/deblocking/smartdeblocking.html

5. Ричардс С., Вудс Р. Цифровая обработка изображений. // М.: Техносфера, 2005. 1072 с.

6. Вестник Национального Технического Университета “Харьковский политехнический институт” Выпуск 114. // Харьков: НТУ “ХПИ”, 2001. 128с.

7. Shalkoff R.J. Digital image processing and computer vision. // – New York-Chichester-Brisbane-Toronto-Singapore: John Wiley & Sons, Inc., 1989. 489 p.

8. Шапиро Л., Стокман Дж. Компьютерное зрение. Пер. с англ. // М.: БИНОМ. Лаборатория знаний,. 2006. 752 с.

9. Проблемы бионики. Всеукраинский межведомственный сборник. Выпуск 50. // Харьков: “ХГТУРЭ”, 1999. 217с.

10. Калинкина Д. Ватолин Д. Проблема подавления шума на изображениях и видео и различные подходы к ее решению [Электронный ресурс] // Калинкина Д. Ватолин Д. - режим доступа: http://cgm.computergraphics.ru/content/view/74

---

[Biography] [Abstract]

---