Pre-Learning-Based Semantic Segmentation for LiDAR Point Cloud Data Using Self-Organized Map

• Published in: Role of Edge Analytics in Sustainable Smart City Development pp. 171 - 188
• Main Authors: Rajathi, K, Sarasu, P
• Format: Book Chapter
• Language: English
• Published: United States John Wiley & Sons, Incorporated 2020; John Wiley & Sons, Inc
• Subjects: environment perception and modelling; LiDAR point cloud; Self‐Organized Map; Semantic segmentation; smart environment; SOM; unsupervised learning
• ISBN: 9781119681281; 1119681286
• DOI: 10.1002/9781119681328.ch9

This chapter presents a framework for reflexive understanding the issues surrounding challenges like smart cities, agricultural environment for autonomous vehicle, and mobile rover's navigation purpose. Reflexive environment perceiving system have many applications like reverse engineering, modelling, autonomous car, autonomous robots, Simultaneous Localization and Mapping (SLAM) vision of navigation. In reflexive perception of environment, sensors play a vital role. Light Detection and Ranging (LiDAR) is an active sensor used in many research applications. LiDAR is remote sensing method used to examine the surface of the earth. Nowadays stunning systematic investigation is going on with LiDAR sensor by dint of its accuracy. It is an active sensor used to detect the object with high accuracy. The main concern in LiDAR is first, visualizing the real time error‐free reflexive environment perception and, secondly is construction of local map for perceived environment. The main issue in modeling the agricultural or smart environment using LiDAR is to make the system to understand and interpret the environment in the right way to the user. Initially, semantic segmentation was developed for the roadside habitat and industrial purposes not for smart cities agricultural environment. To achieve this we propose a new framework and a new pre‐learning process is framed to train the system using SOM clustering algorithm regardless of illumination. Pre‐learning process gives a small knowledge about the environment. This process is implemented and tested with the real time data, the result is analyzed for further rectification.