Research On Tracking Registration Algorithm Of Augmented Reality Based On Feature Points

Augmented reality is a visualization technology that combines virtual information with the real environment to realize interaction,it has developed rapidly in recent years as an emerging technology.Augmented reality technology mainly includes tracking registration technology,virtual and real fusion technology and real-time interaction technology.Especially,tracking registration technology is the key to achieving augmented reality.The existing augmented reality tracking registration technologies mostly adopt the method of collecting feature points to recognize target and performs pose estimation through feature points extraction,matching,and tracking to complete the virtual-real fusion.Aimed at solving the problems such as low operation rate,insufficient robustness and the influence of complex scene,we proposed the improved feature points detection and matching algorithm to improve the efficiency and accuracy of feature points matching in this paper.Based on the above method,we proposed the improved tracking registration methods based on regions and feature points to improve the accuracy,stability and real-time of tracking registration.The main contents of this study are as follows:(1)Due to the problem of high computational complexity and low accuracy of the feature points detection and matching algorithm,this paper proposed the feature points detection and matching based on ORB and FREAK algorithm to improve the efficiency of feature points extraction and matching.To overcome the shortcomings of the RANSAC algorithm to estimate the homography matrix,Hamming distance is used to coarsely filter the mismatched point pairs and arrange them in ascending order.And the method iteratively estimates the optimal homography matrix by sequential sampling.This approach improves the efficiency of iterative calculation and the estimation accuracy of the homography matrix.By using this method,the results show good robustness in the case of rotation,scale,perspective and brightness changes.(2)Aimed at the problems of time-consuming calculation and susceptibility to interference from complex environments of tracking registration method based on feature points detection and matching,this paper proposed a hybrid tracking registration method based on target tracking.The CBWH tracking algorithm is used to track the target,and the method introduces the template matching relocation module and target model update strategy to improve the stability of continuous tracking.By using the improved feature points detection and matching algorithm in the tracking target area for pose estimation,it can still accurately complete tracking registration in complex scenes,and basically meet the real-time requirements of augmented reality.(3)Aimed at the problems of missing feature points and error accumulation during pose tracking of existing tracking registration method based on the Optical Flow algorithm,this paper presents a tracking registration method based on Optical Flow algorithm for feature points selection and pose recovery,which only selects feature points of distance constraints in the target area.The 2D feature points and the 3D coordinate points are calculated one-to-one to perform pose estimation to complete the virtual and real registration.When the target exceeds the boundary,occlusion,etc.,restore the feature points sequence of Optical Flow algorithm and camera pose by feature points matching.Compared with the existing tracking registration method,the result is more stable,robust and real-time under long-term tracking.