Human Sliced Feature-based Head Segmentation on 3D LiDAR Point Cloud Data

Human surveillance can be achieved while reducing extensive privacy concerns by utilizing a 3D LiDAR point cloud data application method. However, due to its lack of organization, point cloud data necessitates additional exertion in processing. Similar to identifying important body components like the human head, segmentation in 3D LiDAR human point clouds presents unique difficulties. This paper aims to propose an alternate method for segmenting human head in 3D LiDAR point cloud data, specifically focusing on walking behavior using the raw KITTI dataset instead of exploiting RGB image data. The dataset used in this study consists of 54 frames of 3D LiDAR point cloud projections capturing human activity as individuals walk toward the LiDAR sensor. Head segmentation can be effectively achieved by utilizing the layer plane features in the vertical axis of human objects. This approach thoroughly analyzes the percentage changes in plane width by utilizing Principal Component Analysis (PCA) in each layered area. When there is a significant percentage difference of α > 25% between layers, specific sections of the human head can be retrieved and utilized for further research in anomaly detection.