Double Hierarchical Feature Matching Strategy Based Pose Estimation Of Object And Its Application

In recent years,with the popularization of augmented reality technology in commercial field,object pose estimation has gradually become a hot research topic.As the basis of augmented reality technology,it not only plays a key role in augmented reality technology,but also in intelligent monitoring,robot motion control and aircraft control.At present,the method based on feature point matching is often used to estimate the object pose.Compared with other methods,it is simpler and more stable.However,it only uses single-layer feature point matching currently,which leads to weak robustness in different environments and can not meet the needs of users in precision and real-time performance.To address the above problems,this paper proposes a novel method of object pose estimation based on double hierarchical feature matching strategy.Two kinds of feature point matching algorithms are used to carry out different combination experiments,which generated two different applications: 3D object pose estimation and multi-cameras frame synchronization based on double hierarchical feature matching strategy.3D object pose estimation based on double hierarchical feature matching strategy is described as follows.Firstly,the shallow feature point matching algorithm is used to relocate the target object and initially estimate the object pose.Then,two different point matching algorithms are used to obtain matching feature point pairs.Finally,all the matching points are used to accurately estimate the target object pose.In the tracking phase,searching the matching feature points only around the target object by the previous frame positioning result can greatly reduce the feature matching computation.In order to improve the matching accuracy,cross-validation,the algorithm of KNN and RANSAC are used to optimize the final matching results.A multi-camera frame synchronization algorithm based on double hierarchical feature matching strategy is proposed to tackle the asynchronized picture frame caused by different cameras simultaneously capturing the same scene.In order to solve the problem that the camera hardware can not accurately obtain the timestamp of the captured image,a synchronous video is designed to represent the timestamp change,meanwhile multiple sets of cameras capture the display of the synchronizedvideo and estimate the position of the display that plays the synchronized video.Then,the relationship between the shooting timestamp and the displayed number is modeled,and the multi-camera frame synchronization is completed by using the constructed mathematical model.Finally,in order to improve the accuracy of object pose estimation,redundant line strategy is added to redesign the synchronous video,and the double hierarchical feature matching strategy is used to estimate the pose of the synchronized video display in each frame.The above two applications are verified by experiments,which separately test the feature point matching performance of SIFT,SURF,ORB,deep point matching and optical flow,and test the feature point matching performance by different combinations.More suitable solution is selected for different applications.The experimental results show that,compared with single hierarchical feature point matching,double hierarchical point matching strategy can obtain more robust feature point matching pairs,which brings a higher precision of estimated object pose.