Research On Feature Extraction Of 3D Scattered Point Clouds

With the development of 3D measurement technology,the digital scanning equipment can effectively obtain the high precision surface model of the physical objects in the real world,and become the main means to obtain the 3D data.Three dimensional point clouds model has been widely used in the field of pattern recognition,3D reconstruction,model segmentation and so on.Feature extraction is the research hotspot of image processing.Therefore,the importance of point clouds data processing is increasing day by day.Based on the summary of feature extraction technology research status at home and abroad,with the Markov Random Field(MRF)model is applied in the field.This thesis from the establishment of typical MRF model and MRF model is established by extracting the feature points is given in research ideas and solution frame.The main contents of this thesis include:Firstly,a global feature extraction algorithm based on Markov Random Field is proposed.The algorithm is based on the classic MRF model.Based on observation point clouds distribution histogram fitting Gauss distribution,according to Bayes estimates for the prior problem is transformed to the maximum a posteriori probability,deduced the solution with the minimum energy of the Random Field,reduction of simplified objective function,solving the function and extracting the feature points.The algorithm for traditional algorithm of problems of manual adjustment of parameters and threshold setting,flexibling integration of MRF model of adaptive,can effectively avoid the disadvantages of the traditional algorithm,and improves the adaptability and efficiency of the algorithm.Secondly,a feature extraction algorithm based on salient feature points is proposed.The core idea of this algorithm is the improvement of the classic MRF model,and the main difference is the establishment of random field model:through significant degree function constructed of point clouds computing the significant degree of scattered points,combinating geodesic distance and saliency to construct Reeb diagram,extracting salient feature points,according to the distance of salient feature points and center point to compute the joint density function for MRF.The rest of thought likes the before.This algorithm avoids setting the initial threshold value and Gauss curve fitting to point clouds of the above algorithm,making the point clouds feature extraction can jump from the traditional thinking about curve fitting and feature parameters of setting.Thirdly,these two feature extraction methods proposed in this thesis are mainly applied to the Virtual restoration project of Terracotta Army of Qin fragments.The experimental results show that these algorithms can effectively extract feature points to the Terracotta Army of fragments,and contrast to the traditional algorithms,the two have adaptive and efficient,which lays the foundation for the subsequent virtual restoration work of Terracotta Army.