Reseach Of The Key Problems On Kinect-based 3D Reconstruction Of Shaft-type Parts

Three-dimensional reconstruction is a technology which builds digital model of the objective world in the computer by utilizing computer digitization approaches. The aim of mechanical parts model reconstruction is finding some form of mathematical description, describe the parts precisely and concisely to the geometrical characteristics of the entity, calculate and analysis of the model, it provide the theoretical basis for such as parts of the processing and manufacturing, virtual simulation and rapid prototyping application, it is one of the research hotspots and difficulties in the areas of virtual reality,reverse engineering and so on.Shaft-type parts is an important part of complex mechanical equipment, it is general by some basic shapes according to certain topoiogical structure combination and location relationships. This paper proposed a solution of three-dimensional reconstruction based on KinectTM depth sensor. Firstly, RGB-D data was captured by Kinect and the depth information was converted into 3D point cloud with actual model by using coordinate transformation; Secondly, the parts point cloud data would be extracted through threshold segmentation algorithm. According to the noise characteristics of point cloud data, a fast bilateral filtering could be used to the data perform denoising operation; An improved point cloud segmentation algorithm was proposed to acquired the point cloud database. Thirdly, based on surface normals and the curvature features of the point cloud, proposed an improved K-means clustering segmentation algorithm, each feature point cloud gathers of shaft-type parts would be segmentated. Besides, proposed 3d mechanical parts recognition framework based on the depth information, learned the 3d point cloud geometric feature description and extracted shaft-type parts recognition and retrieval. Finally, we analyzed 3d point cloud data based on sampling estimation algorithm. The parameters for the point cloud database were analyzed precisely. A completed 3d digital model was reconstructed. It is overcome the problem of the device such as sensor data acquisition low precision, incomplete data, satisfied the requirement of the shaft parts reconstruction.On the basis of the above theoretical analysis, this paper used a Kinect as an input device to captured depth and color data, processed the data and operated algorithm by a typical configuration of computer, a prototype system was build and has carried on the experiment. The results show that the reconstruction system has much better performing recognition, describe and reconstruction of shaft-type parts.