Passive computer vision algorithms often struggle when encountering the real world. If you attempt to perform stereo matching or edge detection on a smoothly curved, uniformly painted white wall, the algorithms fail. Because every pixel looks identical to its neighbor, a camera cannot reliably find corresponding features across different views.
Structured light triangulation solves this problem by shifting from passive observation to active illumination. Instead of relying on ambient light and natural surface textures, a structured light system projects a known geometric pattern—such as a line, a grid, or a sequence of coded stripes—onto the scene. By observing how this known pattern deforms when it strikes a three-dimensional surface, a computer can calculate the exact geometry of the scene with remarkable precision.
The Core Geometry of Active Triangulation
To understand why structured light works, consider the classic geometric principle of triangulation. In a standard passive stereo setup, two cameras are separated by a known distance (the baseline). To find the 3D coordinates of a point in space, the system must locate that exact point in both camera images, trace the optical rays backward, and find their intersection.
Structured light replaces one of these cameras with a light projector. Because the projector's spatial position and the exact geometry of the light it emits are known ahead of time, the system only needs to find the pattern in a single camera image.
[Projector] . . . . . . . Baseline (Known) . . . . . . . [Camera]
\ /
\ /
Projected Ray Observed Ray
(Known Angle α) (Observed Angle β)
\ /
\ /
`------------------> [Surface Point] <-------'
(Calculated 3D)
In the earliest implementations of this concept, such as the foundational work by Shirai et al. and Agin et al., the projected pattern was a single line of laser light. When a flat sheet of light intersects a curved object, the line appears deformed from the perspective of an offset camera. Because the camera's line-of-sight vector and the projector's light plane are fixed, calculating the 3D coordinate is reduced to finding the mathematical intersection between a line and a plane. As demonstrated by Roecker et al. and Yamashita et al., this fixed spatial relationship allows systems to map 2D image coordinates directly to 3D space without complex feature matching.
Solving Correspondence with Coded Light
Projecting a single line requires physically sweeping the laser across an object to map its entire surface. To capture a dense 3D shape in a single snapshot, we can instead project a pattern containing multiple lines or a full grid, a technique pioneered by Will et al.. However, this introduces a new challenge known as the correspondence problem: if the camera sees a hundred lines on an object's surface, how does it know which observed line corresponds to which projected line?
Modern systems solve this by "coding" the light pattern so that every region of the projection is uniquely identifiable. There are two primary ways to achieve this:
1. Temporal Coding (Sequential Patterns)
The system projects a sequence of different patterns over time. For example, Lee et al. utilize Gray-coded binary patterns, where the projector flashes a series of increasingly fine black-and-white stripes. Over the course of the sequence, each point on the surface experiences a unique binary code of illumination (e.g., light-dark-light-light). By decoding this bit sequence, the system determines the exact projector pixel column responsible for illuminating that point.
2. Spatial and Color Coding (Single-Shot Patterns)
When the target object is moving, flashing multiple patterns sequentially causes motion blur and registration errors. In these scenarios, systems use spatial or color variations to encode uniqueness into a single frame. Trucco et al. demonstrate this by projecting a single pattern of multi-colored stripes to capture the dynamic shape of a human foot in motion. Because the local sequence of colors is mathematically unique, the camera can decode the entire surface from a single image.
By bypassing the need for natural visual features, coded structured light serves as a highly reliable method for generating dense 3D point clouds on uniform or featureless surfaces. This makes it an essential tool for high-fidelity applications, such as creating reference data for multi-view stereopsis benchmarks (see Large Scale Multi-view Stereopsis Evaluation), digitizing museum artifacts (Zheng et al.), filling missing gaps in cultural heritage models (Dellepiane et al.), and collecting precise clinical facial dimensions in remote environments (Amirav et al.).
Shrinking the Footprint: Micro-Baseline Systems
Traditionally, structured light systems require a wide baseline—a significant physical distance between the projector and the camera—to achieve high depth resolution. If the camera and projector are too close together, the triangulation angles become shallow, and the lateral displacement of the pattern becomes too small to measure accurately. This requirement has historically prevented structured light from being integrated into compact consumer devices like smartphones or endoscopes.
To overcome this limitation, Saragadam et al. introduced Micro-Baseline Structured Light (MSL). They recognized that when the baseline is extremely small, the complex, non-linear equations governing 3D reconstruction can be mathematically linearized because the resulting pixel shifts are tiny.
To make this linearization robust against noise, the authors designed a specific periodic symmetric triangle pattern. Because a triangle wave has a constant derivative across its entire projection, the relationship between depth and intensity remains perfectly linear and predictable. This allows the system to compute highly accurate depth maps using minimal computational power, even when the projector and camera are separated by only a few millimeters.
Traditional Wide Baseline:
[Projector] <------------------ Wide Distance ------------------> [Camera]
\ /
\______________________ Large Angle __________________________/
Micro-Baseline (MSL):
[Proj][Cam]
\ / Shallow Angle (Linearized via continuous triangle patterns)
Taking this concept further into low-light environments, Wang et al. developed a "Patterned Flash" technique. Standard camera flashes lose energy rapidly over long distances due to the inverse-square law ($1/r^2$). By concentrating all of the flash's optical power into a sparse array of high-intensity dots and separating the flash from the camera by a micro-baseline, the system uses the tiny geometric displacements of those dots to simultaneously reconstruct an image and infer depth in near-total darkness.
Alternative Geometries and Space-Time Trade-offs
The core insight of structured light—modulating illumination to encode spatial information—can be adapted to optimize for speed, computation, or specific scanning environments.
Temporal Super-Resolution
Instead of using spatial patterns to resolve static shapes, Shiba et al. project high-frequency structured light patterns at a frame rate much higher than the camera's shutter speed. This intentionally creates a blurred image where the object's high-speed motion is encoded directly into the blur pattern. By matching these intra-frame patterns against a pre-recorded reference database of depths, the system can reconstruct ultra-fast 3D shape changes from a single standard camera frame.
Light Curtains
In applications like autonomous driving or robotics, mapping an entire 3D volume can be computationally wasteful. The authors of Programmable Triangulation Light Curtains flip the structured light paradigm by synchronizing a steerable line laser with a line-scan camera.
Instead of illuminating a scene and calculating where the light lands, they program the laser and camera mirrors to intersect only along a specific, user-defined 3D surface in space (a "light curtain"). If an object breaks this virtual shell, the camera instantly detects the reflection. This approach cuts out the heavy computational pipeline of full 3D scanning, providing real-time proximity detection that remains robust in challenging conditions like thick fog or bright sunlight.
Key Foundational Work
Grid Coding: A Preprocessing Technique for Robot and Machine Vision (1971) — contributed 2 concept elements to this direction.
Recognition of polyhedrons with a range finder (1971) — contributed 1 concept elements to this direction.
Computer Description of Curved Objects (1973) — contributed 1 concept elements to this direction.