Structured Light

The core principle relies on optical triangulation between a projector, a camera, and the object surface. A projector emits a known pattern (e.g., fringes, grids, or dots) at a fixed angle relative to a calibrated camera. The camera captures the pattern deformed by the object's surface topography. By analyzing the pixel-wise disparity between the known projected pattern and the observed deformed pattern, the system calculates the exact depth (Z-height) for every pixel using geometric triangulation. This requires precise extrinsic calibration between the projector and camera to establish the baseline distance and relative orientation.

The projected pattern dictates the spatial resolution and robustness of the scan. Common strategies include:

• Binary Code Patterns: A sequence of black-and-white stripes projected in rapid succession, encoding pixel columns with unique binary identifiers. Highly robust to ambient light.
• Gray Code Patterns: An enhancement of binary coding where adjacent stripes differ by only one bit, reducing decoding errors at stripe boundaries.
• Phase-Shifted Fringes: Sinusoidal intensity patterns shifted multiple times to calculate a wrapped phase map, which is then unwrapped to yield continuous, high-resolution depth data. This provides sub-pixel accuracy.
• Hybrid Methods: Combining Gray code for absolute phase unwrapping with phase shifting for fine detail, achieving both robustness and high precision.

Unlike time-of-flight sensors, structured light systems achieve micron-level depth resolution at close working distances. By analyzing the continuous phase of a sinusoidal fringe pattern rather than discrete pixel edges, the system can interpolate depth values to a fraction of a camera pixel. This sub-pixel interpolation enables the detection of subtle surface defects such as scratches, dents, and porosity that have minimal contrast in 2D images but measurable depth variation. The depth resolution is a function of the triangulation angle, camera resolution, and fringe period.

The raw output of a structured light scan is a dense 3D point cloud—a set of data points in a Cartesian coordinate system (X, Y, Z). The generation pipeline involves:

• Phase Unwrapping: Converting the wrapped phase map into a continuous, absolute phase map.
• Triangulation: Applying the calibrated system geometry to convert pixel coordinates and phase values into real-world 3D coordinates.
• Point Cloud Filtering: Removing outliers, smoothing noise, and filling small occluded holes using statistical filters. This dense point cloud (often millions of points per scan) serves as the input for downstream metrology, defect detection, and digital twin alignment.

A single projector-camera pair suffers from occlusion shadows—regions where the projected pattern is blocked by the object's own geometry and cannot be seen by the camera. To mitigate this, industrial systems employ multi-view sensor fusion, using multiple projector-camera pairs positioned at different angles. The resulting point clouds from each view are registered into a common coordinate system using Iterative Closest Point (ICP) algorithms, producing a complete 3D model with minimal data voids. This is critical for inspecting complex geometries like engine blocks or turbine blades.

Factory-floor lighting can severely degrade passive stereo vision systems. Structured light actively projects a high-intensity pattern, often in the blue or near-ultraviolet spectrum (e.g., 450 nm), combined with a matched narrow bandpass filter on the camera lens. This spectral isolation blocks the majority of ambient factory lighting (which is broadband or in the visible red/green spectrum), ensuring the camera captures only the projected pattern. This optical filtering provides robust performance without requiring shrouding or darkness, enabling integration directly on moving conveyor belts.

The process of estimating a camera's intrinsic parameters (focal length, optical center, lens distortion coefficients) and extrinsic parameters (position and orientation in world coordinates). For structured light systems, precise calibration establishes the geometric relationship between the projector and camera, enabling accurate triangulation. Without it, depth measurements suffer from systematic errors.

• Removes radial and tangential lens distortion
• Computes the essential and fundamental matrices
• Uses calibration targets like checkerboards or circle grids
• Zhang's method is the most widely adopted approach

The geometric principle by which a 3D point's coordinates are recovered from its projections in two or more 2D images. In structured light, one 'camera' is replaced by a projector that casts a known pattern. The deformation of that pattern, observed by the camera, creates a virtual stereo pair. The intersection of the camera ray and the projector ray determines the depth.

• Active triangulation replaces one passive camera with a projector
• Solves the correspondence problem via coded patterns
• Accuracy depends on baseline distance and calibration quality
• Forms the mathematical core of all structured light depth computation

A fringe projection technique where a series of sinusoidal patterns, each shifted by a known phase increment, are projected onto the surface. The camera captures the distorted patterns, and the phase value at each pixel is computed from the intensity variations across the sequence. This yields a continuous, high-resolution phase map that is directly proportional to surface height.

• Typically uses 3 to 16 phase-shifted images
• Provides pixel-level independent depth computation
• Robust to ambient lighting and surface texture variations
• Unwrapping resolves the 2π phase ambiguity

A collection of discrete 3D data points in space, each defined by X, Y, Z coordinates, generated by the structured light reconstruction pipeline. Each point represents a sampled location on the object's surface. Point clouds are the raw output of most 3D scanning systems and serve as the input for downstream tasks like surface fitting, volume calculation, and defect detection.

• Can contain millions to billions of points per scan
• Often includes additional attributes like RGB color or normal vectors
• Stored in formats such as PLY, PCD, or LAS
• Requires filtering and registration for multi-view alignment

A 2D raster image where each pixel's intensity value encodes the distance from the camera to the corresponding point on the object's surface, rather than color or brightness. It is the direct output of stereo matching or structured light decoding. Depth maps are memory-efficient and align pixel-to-pixel with the camera's RGB image, simplifying sensor fusion.

• Values typically stored as 16-bit unsigned integers
• Can be converted to a point cloud using camera intrinsics
• Enables rapid 2D convolution for defect analysis
• Often visualized with false-color heatmaps for inspection

A pseudo-random, non-repeating pattern of dots or noise projected onto a surface to solve the correspondence problem in single-shot 3D scanning. Unlike phase shifting, which requires multiple frames, a speckle pattern enables depth computation from a single image pair by matching local windows of texture between the projected pattern and the captured image.

• Enables real-time, motion-robust 3D capture
• Used in consumer depth sensors like Microsoft Kinect v1
• Matching relies on block-based correlation algorithms
• Trades spatial resolution for temporal speed