Learning-based 3D LiDAR point cloud homogeneous spatial region segmentation

Nowadays, three-dimensional (3D) Light detection and ranging (LiDAR) point cloud data (PCD) plays a vital role in 3D modeling and autonomous driving environmental perception systems. However, human intervention is needed to handle the massive amount of 3D PCD and to identify the individual objects for decision-making in real-time applications. This manual work takes more time for the identification process and consumes more labor power. To address this challenge, the proposed work designs a learning-based algorithm to segment the homogeneous spatial region of 3D LiDAR PCD. This proposed work consists of two main steps. First, the raw massive PCD has been preprocessed by the statistical-based technique to remove the noise and outliers in the dataset. Second, the preprocessed PCD was segmented into various homogeneous regions based on spatial (geometry) information by using learning algorithm. This presented work is implemented on three different types of 3D LiDAR PCD, and the experimental results show that the given algorithm more accurately segments the homogeneous region of the PCD than the other existing segmentation algorithms such as convolutional model and transfer learning with various algorithms.