A Survey on 3D Point Cloud Compression Using Machine Learning Approaches
Reetu Hooda, Weitao Pan, Tamseel Mahmood Syed
Abstract
Machine learning has been widely used for solving several data processing tasks and recently found applications in data compression domain as well, notably for point cloud (PC). Compression techniques based on deep learning (DL) methods such as convolutional neural network (CNN) have enabled exploiting higher dimensional correlations for improved performance. The most common DL based choice for point cloud compression (PCC) is an autoencoder, while there are few implementations that use recurrent neural network (RNN) and fully connected neural network. This paper surveys the on-going research on PCC using ML approaches. The benchmark datasets with performance metrics are also included. The survey shows the machine learning based methods offer performance comparable to conventional coding methods, while point out directions of promising improvements in the future.