FedGeoDay 2025 | Lightning Talk - Shane Brennan

Shane Brennan discusses a method for classifying 3D point clouds by leveraging 2D computer vision techniques. He focuses on point cloud data collected from moving vehicles equipped with LiDAR sensors, explaining that this data can be represented as panoramic images containing information like range. The core idea is to apply existing 2D object detection and segmentation models, such as YOLO, to these panoramic images to identify and classify objects at the pixel level. Because the pixel location in the panoramic image can be accurately mapped back to the original 3D point cloud using the range and angular measurements, the 2D classification can be efficiently propagated to the corresponding 3D points. Shane highlights that this approach is particularly effective when integrated with Simultaneous Localization and Mapping (SLAM) processes, allowing for classification while mapping the environment. He notes that this method offers an advantage over traditional 3D point cloud deep learning techniques, which often require large amounts of difficult-to-produce labeled 3D training data. • Discusses a method for classifying 3D point clouds using 2D computer vision techniques. • Focuses on point cloud data acquired from moving vehicles equipped with LiDAR sensors. • Explains that LiDAR data can be represented as panoramic images containing useful information like range. • Proposes applying 2D object detection and segmentation models (like YOLO) to these panoramic images to classify pixels. • Highlights the ability to propagate 2D pixel classifications back to the original 3D point cloud using range and angular measurements. • Notes that this method is particularly useful in the context of Simultaneous Localization and Mapping (SLAM). • Presents this approach as an advantage over 3D deep learning methods that require difficult-to-produce labeled 3D data. • Mentions the use of a structured LiDAR sensor like the Ouster OS1. For more content like this check out www.projectgeospatial.com #FedGeoDay #Geospatial #PointClouds #LiDAR #ComputerVision #MachineLearning #AI #YOLO #SLAM #3DData #ObjectDetection #Segmentation