Research Of 3D Object Detection Method In Augmented Reality

Augmented reality(AR)technology is to superimpose the computer-generated virtual information to the real environment to achieve the augmented effect.At present,Augmented reality has been rapidly developed and applied in many industries such as education and training,industrial inspection,gaming and entertainment.In order to enable students to better observe the relationship between the building and the surrounding environment in the architectural design teaching,this thesis introduced augmented reality technology to the design verification section.Students can replace an actual building with a designed 3D model and interact with it in the drone aerial video by using the augmented reality technology,so that the design result display in the architectural design course can be integrated with the surrounding environment,and achieve a better teaching effect.Augmented reality applications must understand the surrounding environment.The key technology is to detect and estimate the pose of the 3D object in reality.3D object detection refers to the detection of specific 3D objects in the image and gives the spatial information of the objects.At present,the 3D object detection in augmented reality is still in the initial stage of development.Most augmented reality systems only detect simple planes and specific images.In this thesis,we deeply study various 3D object detection methods in the field of augmented reality,improve the existing 3D object detection algorithm,and demonstrate the improved effect through comparative experiments.Then,we build an augmented reality prototype system with SLAM and 3D object detection,and it is verified on the drone aerial photographing building data,which is a useful attempt to apply augmented reality technology to architectural design education.The innovations and main work of this thesis are as follows:(1)A method for generating 3D Bounding Box proposals by using the detection data of the previous frame is proposed.On the existing method of 3D object detection based on 2D Bounding Box and vanishing points,we estimate the 3D Bounding Box in the current frame by saving the rotation data of object in the previous frame.It solves the disadvantage of requiring a large number of object rotation samples for each frame and reduces the complexity of system operation.(2)An evaluation convergence method including distance error,shape error,angle error and vertex shift error is proposed.The value of the error function weight is determined through comparative experiments,and the experiments on the public dataset prove the effectiveness of the evaluation method.(3)Based on the ROS platform,an augmented reality prototype system combining3 D object detection and SLAM was established.The effectiveness of the system is verified on the building video of the drone and the virtually generated scene data.It can detect the 3D Bounding Box of the building and realize the augmented reality effect.