Segmentation of Lidar Point Cloud Data using SOM

Sensors play an important role in detecting and perceiving the environment. Many applications of designing a system to understand the environment semantically include vision for navigation, reverse engineering, Simultaneous Localization and Mapping (SLAM), modelling, autonomous vehicles, change detection, and autonomous robots. Accuracy concern, the Lidar sensor is now used in stunning systematic investigations. Lidar sensors generate data in various file formats such as LAS and point cloud data from the sensed environment. The ability to form simple various applications with error-free environment perception from the Lidar point cloud is critical. Navigation of an autonomous vehicle without colliding with obstacles and locating the vehicle on its own may be a difficult task, which is commonly referred to as SLAM. Creating an accurate map for local features in SLAM problems can be a time-consuming process. Many studies are concerned with the creation of a local map for a structured environment. This study will look into the implementation of a self-organized map for an unknown environment. The self-organized map (SOM) learns about the environment and acquires semantic knowledge from a local expert. This is taken as an input and further segments the region, especially unstructured environments like agricultural land, nonindustrial environments. Heretofore, the developed works for clustering and segmentation only for the roadside habitat and industrial purpose. In this paper, we implemented the real-time SOM to perceive any world environment regardless of illumination and is experimented. Implementation outputs are analysed for further rectification.