Enhanced Human Segmentation from 2D LiDAR Data Using DBSCAN Clustering

This study presents a comprehensive approach to segmenting human objects in point cloud data generated by 2D LiDAR sensors within indoor environments, where challenges such as shape variability, noise, and the presence of non-human objects complicate the process. The proposed method leverages the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm, which is well-suited for clustering data points based on density without requiring a predefined number of clusters. The experimental results identified three distinct clusters: Cluster 0 exhibited a regular spread of data points, Cluster 1 featured concentrated mean_x and mean_y values, and Cluster 2 displayed a higher mean_y with a larger standard deviation, reflecting a more dispersed cluster. The DBSCAN algorithm demonstrated its effectiveness with a silhouette score of 0.700 and a Davies-Bouldin index of 0.439, outperforming traditional methods like k-means in terms of clustering accuracy. Human objects were accurately identified, with objects 1 and 3 correctly clustered, while object 2, although a round chair, was grouped with human objects due to feature similarity. These findings underscore the method's potential for improving indoor environment monitoring and autonomous navigation systems, where precise segmentation of human objects is crucial