Bobkov A.V. - Image registration in the real time applications, страница 28

It was found that time complexity of the method is only O(n2), that isbetter than known methods can provide. Low computation complexity makes themethod very useful in processing large images.The accuracy and reliability of the method was also analysed. It was foundthat accuracy of the method is relatively high and allows method to be used onpractice.

The method shows itself to such influences as image brightness variationsand presence of noise, which is a problem for a many popular methods. However,new factors affected to the reliability and accuracy was found: unaccountedgeometric distortions and changing of image content.In such a way, this research presents a novel image registration method thathas a good ratio of primary characteristics: speed, accuracy and reliability. Thesecharacteristics allow method to be used in the applications where both speed andreliability of image registration play an important role. There are, first of all, theapplications related with orientation using visual information in real time.The proposed method itself or its components can be also used in differentareas of computer vision and image processing, since they have goodcharacteristics.

Theoretical results like introduction of extended HT or using SHT129for non-point feature matching can extent existing theory of image analysis andcan find an application in the different areas of this theory.Further researchEach of parts of the present research – contour extraction, feature detection,and image registration – contains both unsolved tasks and new ways for thetheoretical research and practice. So the further additional and deeper research issupposed.Contour detectionThe theory of contour detection is relatively well developed, and it is uneasyto find here something fundamental.

However, there are set of tasks which havepractical importance that must be analysed. There are tasks of image pre-filtration,further development of non-maximum suppression approach, and research ofstatistical detection characteristics.Feature detectionThe key direction of future work here is further development of ExtendedHT concept. This is an integration of EHT approach with other known featuredetection approaches, and application of EHT for the more complex featuredetection (arcs, ellipses, polygons etc).One more separate direction in this area is development of more reliable andeffective algorithm of line merging.Image registrationThe primary problem of proposed method that does not solved in the currentwork is a problem of search space expansion.

It is often required on the practice tofind not only translation but also rotation and scale of one image in relation toanother and even more complex transformations. Direct accounting of these factorsaccording to proposed scheme will lead, as noted, to significant memoryrequirements and loss of accuracy and reliability. Therefore, new approaches mustbe found and tested. These approaches can include different ways of accumulatorarray organisation (e.g. hierarchical search schemes, hashing, projection and theircombinations) and methods of reply cluster analysis.Another interesting direction is investigation of combining of proposedmethod with other image registration methods with different properties anddifferent feature used, in order to characteristics amplification and raising ofreliability.

The features that can be used in addition to line segments can be largespots and specific contour elements. They can be used when line detectionprogress is impossible due to environment influence or image character. These130image registration methods itself are well researched, however, usage of theircombinations are separate important and complex task.Analysis of image registration method in the physical environmentFirst stage of analysis is research of proposed methods on the dynamicalcomplex of semi-physical modelling. The suitable tool for this purpose is thedynamical modelling complex based on robot-manipulator PM-01 (PUMA-560type).

This stand was developed on the Automatic Control Department of BaumanMSTU, Moscow for researching the automated control system componentbehaviour in conditions close to the real environment.Modern systems of control and data processing consist of many componentsthat must interact in the complex and changing environment conditions andwithout operator’s help. It is almost impossible to imitate a whole complex ofconditions that affected to the separate components.

It is often too expensive toproduce and lunch working prototype of the system. The solution is to use semiphysical modelling. It allows keeping most significant information links betweencomponents and to research influences that will affect to the system in the realenvironment.The image analysis system will be affected by different factors:- Variation of the image characteristics due to the plant motion;- Variation of observe conditions due to daily, season and weather changingof lightning conditions;- Raster corruption due to mechanical influences to the camera and vibrations;- Presence of noise both on image data and synchronization signals in theimage receiver and influence of electrical influences.- Geometrical distortions due to non-linear properties of optical system and 3dimentional character of the image;- Asynchronous character of processes of image obtaining, analysis, decisionacceptance and plant control.The structure of the dynamical modelling stand is described in Appendix A.Next stage is research of image registration methods on the base of spaceimages of earth surface obtained from small-class satellite “Baumanets”.

Itcontains four-channel video camera that operates in the visual and near infraredspectrum diapason. On the target trajectory, camera can provide 4096-pixel widthstripe of image data with resolution of 50 meters per pixel. Camera transceiver cansend obtained images to the fly control centre for its further analysis. The videosystem will be used in the task of ecological situation monitoring for the areawhere the trajectory is going over. Another task is research different variants ofposition determination methods on the base of comparison of earth images with thestored map.

The lunch of satellite is planned on February 2006.Research of position determination methods is performed by stationary toolsof the fly control centre. The desired result of this part of work is conclusion aboutsuitability of proposed methods for position determination tasks, and requirementsto the aboard hardware for implementation of these methods.131Related publications1.Бобков А.В. Выделение отрезков на основе гистограмм направленности взадаче анализа видеосцены по последовательности изображений.Электромагнитные волны и электронные системы, №8 т.7, 2002г.(Bobkov A.V.

Line segment extraction based on direction histogram in the taskof video scene analysis of video sequence. Electromagnetic waves andelectronic devices, Moscow, 8(7) 2002)2.Бобков А.В., Комарцова Л.Г. Восстановление функциональной зависимостипо эмпирическим данным ассоциативной нейронной сетью. Методыисследования и проектирования сложных технических систем. Труды.МГТУ им. Н.Э. Баумана, Вып.575, 1999.(Bobkov A.V, Komartcova L.G. Restoration of functions on experimental datausing associative neural network. Methods of development and investigation ofcomplex technical system. Papers of Bauman MSTU, 575, 1999)3.Бобков А.В.

Система ориентации по видимому изображению местности наоснове анализа набора линий. Сборник докладов IV конференции молодыхученых «Навигация и управление движением». Санкт-Петербург, 2002,С.137-141(Bobkov A.V. Orientation by earth surface images based on line set analysis.Papers of IV conference of young scientists “Navigation and motion control”,St.Pitersburg, 2002.– pp.137-141)4.Бобков А.В.

Выделение отрезков на изображении в задаче ориентации повизуальной информации. Вестник МГТУ. Приборостроение, №3(48) 2002(Bobkov A.V. Line segment extraction on the image in a task of orientation byvisual information. Vestnik MGTU, Priborostroenie, №3(48), Moscow, 2002)5.Бобков А.В. Интегрированная система анализа визуальной информации.Труды симпозиума INTELS’2002, Калуга, 2002, С.225-227(Bobkov A.V. Integrated system for visual information analysis.

Papers ofINTELS’2002 international symposium, Kaluga, 2002, pp.225-227)6.Бобков А.В. Модельный мир: философия наблюдателя. Материалымеждународной научной конференции «от истории природы к историиобщества и будущему человечества». Секция «Информати-ка, экономика,искусственный интеллект», М.,2002, Часть 2, С.10-14(Bobkov A.V. Model world: observer’ philosophy. Papers of internationalconference “from the nature history to the history of society and human future”,section “computer science, economics and artificial intelligence”, part 2,Moscow, 2002, pp.10-14)7.Бобков А.В.

Адаптивная система навигации по видеоизображению. Труды6-го международного симпозиума INTELS’2004, Саратов, 2004, с.225-228(Bobkov A.V. Adaptive navigation system based on image registration. Proc. of6th International symposium INTELS’2004, Saratov,2004, pp.225-228)132ReferencesAtiquzzaman M. and Akhtar M.W. (1994) Complete line segmentdescription using the Hough Transform, Image and Vision Computing 12(5)pp.267-273Ballard D.H.