Lidar Point Cloud Compression, Processing and Learning for Autonomous Driving

Abstract

As technology advances, cities are getting smarter. Smart mobility is the key element in smart cities and Autonomous Driving (AV) are an essential part of smart mobility. However, the vulnerability of unmanned vehicles can also affect the value of life and human safety. In this paper, we provide a comprehensive analysis of 3D Point-Cloud (3DPC) processing and learning in terms of development, advancement, and performance for the AV system. 3DPC has recently attracted growing interest due to its extensive applications, such as autonomous driving, computer vision, and robotics. Light Detection and Ranging Sensors (LiDAR) is one of the most significant sensors in AV, which collects 3DPC that can accurately capture the outer surfaces of scenes and objects. Learning and processing tools in the 3DPC are essential for creating maps, perceptions, and localization devices in AV. The intention behind 3DPC learning and practical processing tools is to be considered the most essential modules to create, locate, and perceive maps in an AV system. The goal of the study is to know ``what has been tested in AV system so far and what is necessary to make it safer and more practical in AV system.'' We also provide insights into the necessary open problems that are required to be resolved in the future.