Research On Visual Pose Estimation Technology Based On Augmented Reality

Augmented reality is a technology developed on the basis of virtual reality.Its purpose is to accurately superimpose computer-generated virtual objects into the real scene and achieve seamless integration of the real and virtual scenes,so as to complete the enhancement of the real scene,thus assisting the user in understanding the real world,Therefore,the registration of virtual reality scene is the key to augmented reality technology,and the most important link between virtual reality registration technology is posture estimation technology.Compared with pose estimation methods based on sensors such as mechanical devices,ultrasonic devices,magnetic sensors,inertial devices and GPS,the method of pose estimation based on vision has the advantages of low cost,easy operation and more application scenarios,once became the hot point of academic research.In this thesis,based on the augmented reality application background,a pose-based estimation technique based on visual method is studied.The main contents are as follows:1.The feature detection and recognition based on artificial markers are realized,and the four corner points of the square markers are extracted as feature points.In the original ARtoolKit detection methods,the Hough transformation algorithm and Huber loss minimization algorithm are introduced to improve the original detection method,so as to reduce the effect of outliers on the fitting result and improve the robustness of the result.2.The accuracy and robustness of the DLT algorithm used by ARtoolKit are improved.In this thesis,the objective function is redesigned to improve the jitter problem of registering virtual objects due to insufficient resolution of the DLT algorithm.3.In order to improve the method of solving objective function,this thesis designed a comparative experiment,the artificial marker placed on the electric rotary table,the camera is fixed above the turntable.Acquisition of image frames with artificial markers,using a variety of optimization algorithms including Gauss-Newton algorithm,Levenberg-Maquard algorithm,Dog-Leg algorithm.Among them,Dog-Leg algrothem works best,the experimental results of the reprojection error is 2.303 pixel,better than the original algorithm of ARtoolKit reduced by 54.4%.4.For the augmented reality application environment,a system interface of realtime rendering camera pose trajectory is designed.The system can capture image frames with markers,detect the features of manual markers and identify ID,and use the non-linear optimization method to accurately and robustly estimate the pose of the camera.Real-time augmented reality registration,while achieving the system interface in real-time rendering camera pose trajectory.