AR Spatial Structure Recognition via Point Cloud Normal Vectors

The patent segments the spatial recognition process into multiple independent modules: point cloud extraction, normal vector calculation, plane segmentation, representative plane selection, and spatial structure extraction. Each module processes specific aspects of the spatial data independently, allowing the system to achieve high accuracy without relying on pre-learned features by breaking down the complex recognition task into manageable segments that can be processed systematically.

The patent enables the system to perform self-service spatial recognition by automatically extracting geometric features directly from the captured image data without requiring external markers or pre-placed reference objects. The point cloud extraction and normal vector calculation modules process the spatial environment autonomously, allowing the system to adapt to any space regardless of whether markers are present, thereby achieving both reliability and versatility.

The patent transitions from 2D image plane analysis to 3D spatial structure recognition by extracting point clouds and calculating normal vectors in three-dimensional space. This dimensional transformation allows the system to capture the full geometric complexity of the environment, achieving high accuracy in 3D posture information while maintaining adaptability to different spaces without markers. The normal vector calculations provide additional dimensional information that enhances precision.

The patent performs preliminary extraction of geometric features (point clouds and normal vectors) from the spatial environment that can be stored and reused for augmented reality applications. By pre-processing and extracting fundamental geometric structures without requiring full pre-learning of the space, the system creates a reusable spatial framework that maintains accuracy while enabling rapid adaptation to new spaces through the same extraction process.