Augmented Reality Scene Generation And Visualization Based On Computer Vision

Augmented reality technology is an emerging computer methodology of using computergraphics and stereo display technologies to achieve the effect of fusing the virtual object withthe real environment. Therefore, two kinds of information (virtual information and the trueinformation) can be combined together for interactive3D multimodal applications such as3Dinteractive virtual display, medical training etc. Specifically, under the premise of the realscene captured by video camera, the information generated by computer can be placed withina real scene to allow the user to experience the virtual information as a part of the real scenewith unique immersive experiences. Augmented reality technology has a large number ofapplications ranging from monuments restoration, entertainment, education and so on. Thisthesis investigates the use of augmented reality technology with computer vision based3Dregistration technologies to restore the building for3D indoor scene and to reproduce thescene of the architectural environment, studying its core technology. The ultimate objective ofthis project is to develop an augmented reality prototype system based on ancient buildings toenable people to better understand the history of ancient China and the Chinese cultureheritage.As ancient buildings have been affected by both natural and manmade damages duringhundred and thousands of years, the task of3D reconstructions of the ancient building’sindoor environment can be quite difficult. This project is targeted to use both computer visionand augmented reality technologies for3D reconstructions of indoor environmental of ancientbuildings, the last registration method be carried out a number of improvements. The first stepis the process of registering based on physical feature of the historic building site byextracting physical features from images obtained from multiple views of the building. Thismethod not only increases the frequency of sifting physical feature, but it also increases thecomplexity of the algorithm. As computational efficiency is a key performance indicator of areconstruction algorithm, the method applied in this project also increases the computational time. In addition, in the process of the system operation, the key frame of video imagesrecorded by a digital camera may sometimes appear as a set of flat image features. Under suchcircumstances, an improved eight-point algorithm for calculating the camera’s fundamentalmatrix appears degenerating. Therefore, in order to extend application algorithm and improvethe efficiency of the reconstruction algorithm,2D feature extraction algorithms that are basedon the composite character and five-point camera estimation that used for estimating themovement of camera are studied.We summarize the main contributions of this dissertation as follows,1. Aiming at the indoor environment register problem of the historic building site,composite character register algorithm is put forward, which is based on extracting bothphysical and signal features. This algorithm solves the problem of requiring a large number ofimages in the process of the indoor environment registering. The method can reduce thecomplexity of the algorithm, thus, improving the operating efficiency of system. This researchproject has developed a new approach to address the problem of3D feature extraction byintroducing the signal feature with the image features to aid the3D feature extractingcomputations with the use of a set of text feature in the experiment. Therefore, the newapproach improves the registration accuracy and the deficiency by using a fusion algorithm.Experiments have shown that new algorithm raised by this text can meet the need of thisapplication.2. Aiming at the current calculation of fundamental matrix of improved eight-pointalgorithm for camera, it appears a mass of flat scenes in the view of camera and then itdegenerates. In order to extend application of algorithm and improve the stability of algorithm,a five-point algorithm is used for calculating the essential matrix of camera to estimate themovement of the camera. It realizes the attitude estimation of the camera. It has been foundthat for the target application of the indoor environment register of the historic building site,the five-point algorithm has greater stability and higher registration accuracy.The experiment results show that the registration method proposed can be successfullyapplied in the indoor environment register of building, it has achieved the desired researchtarget. The registration method that put forward in the thesis can be applied in the indoorenvironment register of the building with the stabile tracking and precise registration. You canquickly create augmented reality scene information by this method. The development oftechnology laid a technology research foundation for the ultimate goal which is to develop theaugmented reality system based on the ancient ruins.