Natural Feature Based Registration Method In Augmented Reality Environment

Since the sixties the development of computer graphics technology has developed rapidly. And in this field virtual reality used successfully in various industries is the predecessor of augmented reality. As augmented reality is more and more applied to industrial, medical, education and other fields to help people work better, their corresponding learning and research will be of great urgency. The registration is one of the key technologies of augmented reality. It is the technology adding computer-generated virtual objects to the real-world of users'perception, and the technology enhancing the human visual system or guiding their behaviors. The fundamental task of registration is to determine the transformation between different coordinate system of the virtual model, and is to display it timely in the correct location.This thesis studies the method to achieve the registration through mapping of character-istics in the plane. First, realizing the registration status in monocular camera for augmented reality systems, We confirm the indoor environment and experimental subjects, using the OpenCV library for video image capture. Briefly we describe the characteristics of the histogram based on single channel color, from which we determine the appropriate threshold through the description of image segmentation to separate the target area from the background. we restrict template searching area by detecting the edge of the image target region.Secondly, we define the criterion for natural feature, and propose the screening method for feature points. The process determines the corner identification in our laboratory conditions here. This thesis describes the coordinate conversion among graphics coordinate system, the object coordinate system, the scene coordinate system and camera coordinate system of the registration. How to register and position is actually trying to solve the details of the location and the orientation of the video captured part in real-time, that is the position and orientation of camera coordinate system relative to the natural feature points coordinates. According to these information it calculates and displays the location and orientation of the virtual object.In the camera's perspective projection transformation, the results show that the transformation between a scene plane and an image plane is a plane projective transformation. Through this principle and the corresponding transformation. we get the transformation matrix of the camera coordinates relative to the world coordinate system analyzing the relationship between three pairs of coordinate transformation. And ultimately complete the registration of virtual objects in the lab environment making used of the feature points in the camera imaging point.Finally, this thesis analyses the experimental environment and the system architecture, completes augmented reality experimental registration system based on monocular camera, and researches the framework of the running processes in registration and analyzes where the registration error is in augmented reality system, relying on graphic tools and computer vision library. Finally it shows that this experiment can better complete the registration, and meets the requirements of accuracy and instantaneity.