A Study On Preprocessing Of 3D Point Cloud Data

3D Laser Scanning Technology has been rapidly growing in recent years. The application of 3D Reconstruction of point cloud data model obtained from scanning device is becoming gradually mature. It is widely used in construction, military, manufacturing, historic relics protection and other fields. Kinect, a Microsoft’s depth somatosensory device, is a kind of scanning device based on optics. For its high accuracy, little affection from light, small size and low cost, it is widely accepted.In the process of collecting data, the obtained huge amounts of data have noise inevitably due to the measurement and random error. So before the 3D Reconstruction, there is a need to preprocess the original point cloud data, including removing of outliers, reducing and smoothing the point cloud.In this paper, Kinect is used to scan the real scene to obtain the disordered and scattered point cloud data. The obtained point cloud data contain noise, such as outliers etc.. If the original data are 3D reconstructed directly, the result will be seriously affected. According to the distribution characteristics of outliers, the method based on bounding boxes is adopted to remove outliers in this paper. In order to improve the efficiency of the following data processing, the bounding box is used to streamline the point cloud according to its uniform distribution. After being scattered, a large amount of redundant points have been deleted with details retained. After these processing, the purpose of noise reduction has been achieved. The smooth processing makes the concave-convex model surface more smooth and the result more realistic. We give three smooth methods in this paper, which are based on the Moving Least Squares algorithm, Gauss Kernel algorithm and Laplace algorithm. The comparison of the three methods is given in the experiment part.