An Algorithm for Key Object Point Cloud Extraction Suitable for Subway Tunnels

In subway tunnel environments where GPS signals cannot be received, autonomous train positioning faces significant challenges. To address this issue, this paper proposes an algorithm that integrates multiple point cloud processing techniques in tunnel scenarios to extract key objects and calculate distances, aiming to utilize critical objects within the tunnel for assisted positioning. Specifically, the algorithm combines techniques such as point cloud filtering, clustering, and plane segmentation. Initially, radar point cloud data is preprocessed using the coordinates and intensity information to remove noise points while retaining essential data. Statistical filtering methods are then applied to further refine the point cloud by setting appropriate parameters to eliminate outliers and enhance data quality. Subsequently, the filtered point cloud data undergoes plane segmentation, where threshold parameters are set to segment the data into multiple planes. Post segmentation, the Minimum Bounding Box (MBB) algorithm is employed to obtain the volume of the smallest bounding box for each plane. These processing steps enable assessment of whether the extracted planes meet requirements, followed by distance calculation using the centroids of compliant planes. Using data recorded from a hardware platform installed on a city subway, the algorithm extracts point clouds of landmark objects within the tunnel and calculates their real distances relative to the train. Experimental results demonstrate that the algorithm effectively extracts landmark point clouds in tunnel environments, thereby enhancing the accuracy of train positioning. This validates the feasibility and effectiveness of the approach in practical applications. The study provides a novel solution for autonomous train positioning in subway tunnels where GPS signals are unavailable.