|
ИСТИНА |
Войти в систему Регистрация |
ИПМех РАН |
||
Overview of methods for estimation the observed velocity of the object in the video streamтезисы доклада
- Авторы: Konovalenko I., Nikolaev D., Kuznetsova E.
- Сборник: УПРАВЛЕНИЕ, ИНФОРМАЦИЯ И ОПТИМИЗАЦИЯ (VI ТМШ) Тезисы докладов Шестой Традиционной всероссийской молодежной летней Школы. Федеральное государственное бюджетное учреждение науки, Институт проблем управления им. В. А. Трапезникова Российской академии наук,
- Тезисы
- Год издания: 2014
- Издательство: Институт проблем управления им. В.А. Трапезникова РАН
- Местоположение издательства: Москва
- Первая страница: 34
- Последняя страница: 35
- Аннотация: The velocity estimation is a special case of the problem of motion estimation and object tracking via the video stream. Several basic approaches of solving this problem are examined in this report.Problem Statement. Two pictures of a moving object with a small interval of time between the moments of the shooting are given. It is necessary to estimate the observed velocity (the displacement in pixels) of the object. Main practical problems: Various noise caused by shadows, illumination, reflections, lighting and background variability. An object does not completely enter the frame. An object is isomorphic. Different parts of the object have different observed rate. A projective transformation, radial distortion of a camera, rotation of the object, differentdistance between a camera and object parts account for this fact.Approaches to solve the problem: The correlation analysis. One should maximize the convolution of two images or their parts for matching them. The separation of the object from its background and from another objects is carried out in this process. The modified algorithm Lucas - Canada. It is based on the idea that the image of the object on the video stream with a constant speed must satisfy the transport equation with the corresponding coefficient. This coefficient by multiple substitutions in this equation recovers statistically. Quite a general approach based on the use of technology of feature points. Refers to the following item: Feature point extraction. Such points are ones that can be found in both images, they are likely to be relevant to each other in terms of the object inthe scene. We need their invariance with respect to any transformation typesof the frame. Feature points are often the extremes of the image, corners, blobs. Calculation of descriptors of feature points. We requires descriptor's invariance with respect to the transformation of the frame, noise immunity,simplicity of calculation. Matching of feature points by minimization of the differences between the relevant descriptors (creating pairs). Estimation of the parameters of the image transformation that transforms one image points into their pairs on the other picture. Having knew the parameters of the transformation, it is easy to identify and find, in particular, the displacement, we are interested in, in pixels.
- Добавил в систему: Коноваленко Иван Андреевич