Based On Filtered Estimate 3D Registration Algorithm Research

Augmented Reality technology is a newly developing advanced technology. As an important branch of virtual reality, and has a incomparable characteristics and superiority. Augmented reality technology generate needed virtual objects or useful tips information through a computer, real-time adding to real scene which the camera gets (the real world of sensory that human eyes can directly observe), which greatly improve the cognitive ability of people to the real world and show more abundant information, also may provide the ability to interact, help people better work in the real word.3D registration is one of the critical technologies of Augment Reality, overlaying virtual objects in real scene exactly, directly affects the availability and stability of the whole augment reality system. The research using computer vision theory and digital image processing algorithm to improve registration becomes hot. This paper introduces a variety of filter technology and the scene natural feature tracking and reconstruction, in view of the unknown environment which augment reality system in using may face, estimate and correcte the camera pose, improve the validity and precision of 3D registration as much as possible, enhance the adaptation and application of the reality system in changeable circumstancesThis paper studies on the work and practice mainly as follows:(1)based on kalman filtering technology registration algorithm of researchkalman filter technology is applied to registration algorithm design.This method according to distinguish the Identification Block in the scene, timely switchs different feature points extraction and matching methods, in view of Identification Blocks maybe partially sheltered, using kalman filter to correcte the camera pose and solve the stability problem that only Identification Blocks must be completely detected ,the AR system can correctly place the virtual objects defects.(2) based on Unscented Kalman Filter technology reconstruction of 3D scene and registration algorithm researchIn order to get rid of excessive dependent on artificial Identification Blocks and when the camera movement range is larger, considering the natural fixed features in the scene, cite the SLAM technology based on Unscented kalman filter to reconstruct the map of the scene , using natural features to calculate the camera pose for 3D registration and the law of motion estimation, get rid of bondage of artificial Identification Blocks , provides a better solution to strength the application of reality system in changeful circumstance.(3)Improved Unscented kalman filter technology for reconstruction of 3D scene and registration algorithm researchWhen the camera is fast moving,using the SLAM technology based on Unscented kalman filter for scene reconstruction, the acquisition of the image of two-dimensional features is not apparent, easy to introduce inaccurate natural features,so that the algorithm should be improved, citing two phase filters.The improved algorithm can effectively handle inaccurate natural features, promote filters fast convergence. When the scene change faster, the sysetm is also able to get accurate three-dimensional position of new joined natural features, greatly improve the system suitability and stability.