Research Of 3D Registration Based On Vision In Augmented Reality

As an extended technology in virtual reality, augmented reality has developed rapidly in recent decades and shows strong value in many fields. It has been applied widely in area of industry, product exhibition, research of medicine, military, education etc. So it becomes a popular research field in the area of virtual reality. 3D registration technique is one of the critical techniques in augmented reality, and it determines the mixing effect. Based on researching on various kinds of 3D registration methods in augmented reality, the works are introduced as two aspects as follows.1. A scheme aiming at markerless augmented reality system proposed.In the methods of 3D registration based on computer vision, this paper demonstrated the principle of registration in details taking the registration method based on fiducial marker for instance. According to that, we proposed the method of realization based on nature features: We used feature detection technique to achieve points coordinates in the image as the input data of tracking, and used the optical flow tracking technique to get matching relationship between two images which was under the situation of the wide baseline, then reconstructed the matching points to get corresponding 3D coordinates, then applied the geometry constraint to getting planar points to construct a local coordinate system, then computed the camera extrinsic parameters, and finally augmented the virtual objects on the image.2. System design, realization and optimization.Based on the scheme above, we realized a prototype system without marker. Aiming at the problem of localization in feature detection, we used existing method to make it better which could make the input data more accurate. Aiming at the problem of similar errors and accumulated errors in tracking, we proposed a tracking strategy combined with feature detection to correct the tracking result frame by frame. As a result of that, the accuracy of system is improved and it also extends the activity scope of the camera. Finally, we evaluated the system at the aspect of real-time, accuracy and robustness.The prototype system was developed on Windows XP platform by Visual C++ combined with OpenCV and OpenGL. The experiment shows that this scheme can meet the basic requirement of application.