RGBD

RGBD refers to data that combines a color image (RGB) with per-pixel depth information (D). An RGB-D image pair provides both appearance and geometric structure of a scene, enabling applications that rely on understanding the three-dimensional layout of environments. The data is typically captured by RGB-D cameras that provide color and depth streams in a synchronized manner.

Depth information is generated using infrared-based sensing technologies such as structured light, time-of-flight, or, in some devices, stereo vision. Early devices like the Microsoft Kinect employed structured light, while newer sensors use infrared time-of-flight or stereo approaches. Depth values are often stored as 16-bit measurements, commonly in millimeters, and aligned to the corresponding color image with a known spatial offset.

To be useful, the depth and color streams must be registered to a common coordinate system. This requires calibration of intrinsics and extrinsics, and often calibration of the relative offset between sensors. In practice, the resolution and field of view may differ between channels, and some depth data may be missing or noisy near object edges or reflective surfaces.

Processing of RGB-D data typically includes depth denoising and hole filling, texture mapping, and the generation of 3D representations such as point clouds, meshes, or voxel grids. Common workflows convert depth to a 3D point cloud using camera intrinsics, then fuse color for textured reconstructions or apply SLAM and 3D mapping techniques.

Applications of RGB-D data span robotics, autonomous navigation, augmented reality, 3D scene understanding, object recognition, and human pose estimation. Benchmarks and datasets such as NYU Depth v2, SUN RGB-D, and ICL-NUIM have driven research and evaluation in the field.