Structured-Light 3D Scanner

Overview

A structured-light 3D scanner projects a known light pattern onto a surface, then images that pattern with two cameras to compute depth. The pattern acts like a reference grid: where the pattern is bright, where it's dark, and how it distorts tells the dual cameras exactly how far away every pixel is. This approach runs fast (5–100 Hz in many systems) and is accurate to millimeters, so it's favored for industrial metrology, quality inspection, and archaeological documentation.

The Projector Unit beams a DLP or laser pattern (often a sequence of stripes, grids, or random speckles) at the scene. The Stereo Camera Pair captures images of the lit surface from two slightly offset viewpoints. The Processing Unit compares the two views of the illuminated pattern, finds corresponding features, and solves for the 3D position of each pixel, traditional stereo vision, but with the pattern providing the texture that raw photogrammetry alone might miss on a blank wall.

For maximum accuracy, Calibration Target Board targets are imaged during factory calibration and in the field, establishing the exact camera intrinsics (focal length, principal point), distortion, and relative positions. Any deviation from this calibration introduces depth error, so careful setup is essential.

How it works

The core idea is stereo vision: if the left and right cameras see a point at pixel $(x_L, y_L)$ and $(x_R, y_R)$ respectively, the disparity $d = x_L - x_R$ is inversely proportional to distance. The Processing Unit scans both images for matching patterns, solves for disparity at every pixel, and converts to 3D coordinates using the known camera geometry (baseline distance and focal lengths).

The challenge is finding matches when the surface is featureless, a blank white wall has no obvious points to track in left and right images. The Illumination System solves this by projecting a known Pattern Light Source pattern. A stripe pattern is simple: vertical light stripes sweep across the scene, and the camera pair imagesenses which stripe fell on each pixel. A checkerboard pattern works similarly. Random speckle (infrared structured noise) is harder to decode but offers higher spatial resolution.

Multi-pattern strategies are common: the projector emits a sequence of different patterns, say, 8–16 frames, and the Processing Unit combines all the constraints to compute a depth map with minimal ambiguity. This takes time (100 ms to 1 sec per full 3D snapshot) but yields dense, accurate geometry.

The Stereo Camera Pair is synchronized to the projector. The Stereo Baseline Bracket determines the depth range and sensitivity: a wider baseline (camera separation) gives better depth accuracy for distant objects but reduces the close-range accuracy. Industrial scanners use baselines of 10–50 cm.

IR Bandpass Filters reject sunlight, which would wash out the infrared pattern. Outdoor use is difficult; most scanners work indoors or in controlled lighting.

Resolution depends on working distance and lens choice. At 0.5 m range with a 50 mm baseline, a 2 MP camera can yield 0.1 mm per-pixel depth resolution, enough for quality inspection of precision parts. At 2 m, the same setup covers a larger area but with ~0.5 mm resolution per pixel.

Post-processing is often essential: filtering noise, filling holes from reflective surfaces or occlusions, and meshing the point cloud into a watertight surface model. Modern systems push this computation to a GPU on the Processing Unit to provide real-time preview.