Abstract
In this paper, we propose a method of intensity-based image matching evaluation to achieve efficient planar object detection with small number of feature points. To detect target object from a camera image, establishing matches between feature points of camera image and those of objects is processed. As correctness in matching procedure effects overall performance of detection task, it is important to determine error-prone matches and filter them out. In previous researches, they have usually focused on probabilistic approaches to figure out error of matches. In their approaches, with large number of matched feature points, it performs optimization process which is iterating estimation of projectivity with randomly selected matched feature points until it finds converging projectivity result. The matched feature points which contribute the projectivity result is assumed to be correct. In these approaches, to acquire reliable result, there should be large number of matched feature points and most of matches should be correct. In this paper, we propose an efficient way to detect the target object with small number of matches. It becomes possible by filtering out wrong matches and increasing reliability of set of matches which is small. To filter out error matches, we exploit intensity information of image. To make intensity information usable, we have devised a method to define geometric structure between feature points and model the intensity information based on it. The experimental results show that it is possible to acquire reliable projectivity with small number of correct matches and enhance overall performance by minimizing the burden for optimization process.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have