The Research On The Point Clouds Registration Method Based On RGB-D Data

3D reconstruction is an important part of computer vision and graphics.3D reconstruction of real objects have a very broad applications in all areas of real life,including the electronics industry,construction,entertainment,digitization of cultural heritage protection,computer-aided medicine,etc.So the research on the point cloud data registration has been the focus of the three-dimensional model reconstruction.Especially with the popularity of 3D scanning devices continue to be introduced,the real object point cloud data access becomes easier,but the data have low-quality,high noise and other shortcomings.The researches of point cloud data processing and registration algorithms have become the key of 3D reconstruction.So the research on point cloud data processing method about the new 3D equipment has a very important significance on the 3D model reconstruction.First,aiming at the problem of the point cloud data scanned by Kinect depth camera with a huge number and more noise,this paper presents the methods of point cloud data simplification and denoising with feature preserving.The algorithms first search the nearest eight points of the target point by K-D tree;Then use the curvature estimation algorithms to approximate the curvature of the target point;Then can realize point cloud data clustering by using the K-means clustering algorithm,and reject noise data depending on the curvature changes of neighboring points and the Euclidean distance of between points;Finally,we simplify the data by the corresponding point cloud simplification method with the feature preserving.Second,this paper presents a RGB-D point cloud data registration method based on the dichotomy iterations.Firstly,we use the calibration object to obtain the relative relationship between the Kinect and rotating platform;Secondly,we extract the feature points from the points to be matched and find the corresponding points of initial points cloud by the curvature characteristics;Thirdly,the dichotomy iterations method isintroduced to find the optimal angle of rotation,in order to meets the minimum of the squared sums of arc distances between the two points cloud.Finally,take the point cloud data that get from several angles to obtain registration under the unified coordinate system and rebuild the 3D model.Finally,the experimental results show that the method is quickly and efficiently by point cloud simplifying and denoising with feature preserving,and for the removal of a large number of external noise has a positive effect.Moreover,the streamline point cloud data have retention the original point cloud features.Illustrations show the registration method of reconstruction object 3D model by using a rotating platform is stable and high efficiency,and also has a significant improvement on splicing accuracy compared with ICP and other methods.