Research On Technologies Of Tracking And Registration For Mobile Augmented Reality

Augmented Reality is a technology that overlays computer-generated 3-D virtual objects onto the 3-D real environment in real time.In the process of overlaying,the virtual objects and real scenes must be properly aligned with each other,otherwise the effect of augmented reality will be reduced greatly.Therefore,the key problem of augmented reality is how to align virtual objects and real environment correctly.The technologies of tracking and registration allow computer devices to get and track its pose relative to real scenes in real time.Then according to the pose,it can properly integrate virtual content with the real scenes.However,mobile devices have to face the resources limitations of small memory size and poor CPU performance.Thus,to solve these problems,this paper proposes some methods to modify the existing tracking and registration algorithm to fit the requirements of mobile devices.The main contents and contributions of this paper are as follows:(1)To improve the timeliness and robustness of image matching algorithm,an improved FREAK feature point matching algorithm(MyFREAK algorithm)is proposed.Firstly,the classic FREAK 8-layer retina model is simplified into a 5-layer one and only 64 receptive field pairs are selected with a greedy search algorithm to preserve efficient pair information as much as possible and reduce the amount of calculation at the same time.And then,a rotation-invariant LBP is designed to encode every receptive filed to increase the discrimination of descriptors.Compared with other algorithms in experiment,the proposed algorithm has the smallest descriptor size.According to the experimental results on Mikolajczy and day-night data set,in most scenes,the proposed algorithm is faster to compute,more accurate to match and more suitable for environment with complex illumination change.(2)This paper uses the error-state Kalman filter to derive the error-estate equations of the preintegration of an inertial system integrating accelerometer and gyrometer readings with bias and noise,using the Hamilton quaternion to represent the orientation in space.The error-state is always small,meaning that all second-order products are negligible.This makes the computation of Jacobians very easy and fast.Compared with the traditional foster preintegration,the ESKF-based preintegration is faster to calculate,and therefore is more suitable for mobile devices.Then based on the OKVIS algorithm,a new tracking and registration algorithm(MyVIO algorithm)is proposed to suit mobile devices.The basic idea of the MyVIO algorithm is to use the ESKF-based preintegration instead of the IMU kinematic model to process IMU data and to apply the proposed MyFREAK algorithm instead of the BRISK algorithm to process image data.In order to improve the robustness of MyVIO algorithm,an visual-inertial initialization method is proposed to bootstrap robustly the tracking and registration algorithm.According to the experimental results on EuRoC data set,in most scenes,the proposed MyVIO algorithm can get higher frame rate and track trajectory more accurately.Finally,based on the proposed MyVIO algorithm,a mobile application is designed and implemented.The mobile application actual operating result shows that MyVIO algorithm can basically satisfy the requirements of mobile devices for timeliness and robustness.