Research On Registration Of Virtual Objects In Augmented Reality

In recent years, people pay high attention to augmented reality technology thatinvolves computer vision, computer graphics and virtual reality and other fields. Itmainly restores the three-dimensional information in the two-dimensional imageinformation and tracks them through the use of computer vision, and then generates thevirtual objects in the video images via the computer. During the process, it is mainlyrelated to the tracking technology and the three-dimensional registration techniques. Thethree-dimensional registration is the key of augmented reality technology. A stablemethod for the registration can make the virtual objects which are generated by thecomputer seamlessly overlay into the video in order to achieve a combination of thevirtual and the real.This paper mainly improves the calibration process of the traditional quadrilateralmanual template. A method for the three-dimensional registration based on the triangularpattern has been proposed. At least the four pairs of corresponding point coordinatesshould be known in the process of traditional registration. In this paper, we used threevertices and barcentric coordinates of the triangle to obtain the four pairs ofcorresponding point coordinates information to compose two orthogonal vectors for theworld coordinate system. According to the four image coordinates of feature points andphysics coordinate value which we have known, we calculate the homograph betweenthe camera image plane and the template plane and deduce the camera extrinsicparameters using the homograph, and then optimize these parameters to get the stableparameters to achieve calibration for tracking. After the success of template matching,we use the Mean-shift method to detect the real-time tracking target. Thus we avoidmatching every frame of the template to reduce processing time and enhance real-timeprocessing capabilities.The method with triangle template has a high recognition rate in the environment of detection process. Compared with the traditional rectangular template, it is more flexibleand efficient. Through experimental testing, the three-dimensional virtual objects can berendered accurately into the physical video image by using this method.