Research On Virtual-real Registration Technique Based On Blob Feature

With extensive application foregrounds in military field, Augment Reality has recently beena hospot of Virtual Reality research. Virtual-reality registration technology is the key issue of anypractical Augment Reality system. The virtual-reality registration technology based on noidentification has the characteristics of low cost and convenience, to which more and moreattention is paid. Focusing on the virtual-reality registration technology based on noidentification, this paper proposes a virtual-reality registration approach based on blob feature.The main work completed is as follows.(1) The virtual-reality registration framework based on blob feature is designed. There aretwo segments in this framework: initialization and online tracking. The initialization phase aimsat the extraction and matching of the blob festures of reference frames; then the camera iscalibrated and the initial map depot is established. The online tracking phase aims at tracking thecamera pose as well as updating the map depot; the virtual-reality superposition in tems of thetracking results is implements.(2) To cope with the high complexity of blob feature extraction algorithm, an efficientapproach for blob feature extraction is proposed. By using the three layer filter function totransform the feature detection into addition and subtraction operaton of the integral image first,the detection efficiency is improved; then by combining the advantages of gradient operationbased on integral and quantization, a gradient operation based on integral-quantization isdesigned and applied to the festure description which improves the efficiency. Experiment showsthat the perfomance of the method proposed is higher than the similar methods.(3) Aiming at the low precision and robustness of the camera self-calibration, this paperpoposes a camera self-calibration method based on layered reconstruction. First we establish thecamera linear imaging model and improve the matrix estimation algorithm based on RANSACby the maximum distance together with the sampling and weighting error. Then theself-calibration problem is transformed into the optimization problem by layered reconstruction.Finally we solve the camera parameters employing the optimization algorithm based on artificialfish swarm.(4) A camera pose tracking approach based on SLAM is proposed. The camera posetracking in this approach is divided into two processes: map building and position tracking. Inthe map building phase, the environment map is constructed using feature points set and keyframes set. Then the L-M algorithm is applied to optimize the map data. In the position trackingphase, a camera motion model is builded and the adaptive camera tracking is realized by pose prediction and reprojection. Experiment shows that the approach can achieve a good effect.(5) On the foundation of the above research, a virtual-reality registration prototype systemis designed and implemented based on blob feature.Finally, the main work is summarized and the development prospects along with theresearch directions are predicted.