JA EN

Point Cloud

A data format consisting of numerous 3D coordinate points (x, y, z) in space. Acquired by LiDAR or stereo cameras, it serves as foundational data for 3D modeling, autonomous driving, and surveying.

A point cloud is a collection of data points in three-dimensional space, representing the external surface of objects or an environment. Each point carries at minimum its 3D coordinates (x, y, z) and may include additional attributes such as color (RGB), surface normal vectors, and reflectance intensity. Unlike meshes or voxels, point clouds are unstructured data with no explicit connectivity between points.

Point clouds are acquired through various sensing technologies. LiDAR (Light Detection and Ranging) measures distances using laser pulse time-of-flight, capturing hundreds of thousands to millions of points per second. Stereo cameras compute 3D coordinates from disparity maps. Structure from Motion (SfM) reconstructs 3D point clouds from multiple 2D images. Depth cameras (ToF, structured light) generate point clouds in real time.

Major libraries for point cloud processing include PCL (C++) and Open3D (Python/C++). In deep learning, PointNet and PointNet++ directly consume raw point clouds for 3D object classification and semantic segmentation. Autonomous driving systems use LiDAR point clouds for real-time 3D object detection of pedestrians, vehicles, and cyclists using architectures like PointPillars and CenterPoint.

Related Terms

Related Articles