Statistical filtering on 3d cloud data points on the CPU-GPU platform

Abstract

Recent advancement in scanning technologies has allowed an object to be represented in the 3D point cloud, which is an effective way to represent the overall view of the data and can be used for many purposes, such in manufacturing and visualization. However, the challenges in handling point cloud data are the noise and massive amount of data. Therefore, this study carries out a denoising process to remove the noise and reduce the size of data using statistical filtering. The process starts with neighboring points calculation using KNN. Then, the points are filtered using the statistical filtering method. This paper used 3D points of Armadillo and Stanford bunny retrieved from Point Clean Net database. To accelerate the performance of the distance calculation in KNN the process is executed on the CPU-GPU algorithm. The results show that the statistical filter has removed an amount of noise and preserved the features of the data. For the developed CPU-GPU platform, it is shown that the efficiency has accelerated the distance calculation process more than 700×.