| Popis: |
The work is devoted to improving the quality of the solving tracking tasks by modern robotic systems. Considering method of tracking objects based on the representation of images in the features form, and using intuitionistic fuzzy logic. Work with images presented in the form of features clouds, due to desire to increase the speed of optical information processing, and the use of intuitionistic fuzzy sets (IFS) is trying to make the behavior of the complex similarity to the thinking of human. Presents the analysis of modern methods of detection features, their description and matching, implemented in the OpenCV library. The comparative characteristics of the considered methods are given. The most suitable methods for the detection and description features on the low quality and noisy images selected. The disadvantages of the considered methods of matching presented. The developing method takes into account the geometric structure of the tracked object. The truth feature matches estimated by categories close to "not absolutely correct", "almost correct ", "absolutely correct ". The rules used for fuzzy inference presented. The developing method divided into three stages. The first is the features detection on the images of the tracked object and the search window. Then compute the description of the neighborhoods founded features to create unique, robust to changes in the rotation and scale descriptors of features. The final stage is a feature matching to determine an unequivocal correspondence between features of the object image and the search window. The results of experiments with the proposed method, as well as estimates of its stability and speed demonstrate the high quality tracking objects in video sequence. The developed method has a high robust to the size of the object changes, its rotation and position in the frame. However, the method is weakly robust to perspective or projective changes of the object, which compensated by timely updating of the object template. In general, this method designed to improve the efficiency of modern robotic systems. |
