Photometric Stereo

Keeping the camera fixed and lighting a surface from several directions, then recovering its shape from how the shading changes.

Photometric stereo was proposed by Woodham in 1980. The camera stays in a fixed position while the object is lit from at least three known directions; under the assumption of Lambertian reflectance (a surface that scatters light evenly in every direction), each pixel’s brightness depends only on the surface normal and the lighting direction, so solving a small linear equation recovers the normal at that point, and integrating the normals yields a depth map. It can recover very fine surface texture, but performs poorly on metallic, glossy, or transparent surfaces. In embodied AI, its most important use is inside vision-based tactile sensors: GelSight illuminates a soft elastic gel surface with red, green, and blue LEDs from different directions, and a camera behind the gel takes a single image; a calibrated lookup table then converts the colors into surface normals, reconstructing the 3D shape of whatever is touching the gel.

ExampleA GelSight fingertip is pressed onto a coin; the height map reconstructed by photometric stereo shows the coin’s embossed relief pattern.

Related

Vision-Based Tactile Sensor · GelSight · Tactile Image · Surface Normal Estimation · DIGIT · Taxim: An Example-based Simulation Model for GelSight Tactile Sensors

Sources

Photometric stereo - Wikipedia
GelSight: High-Resolution Robot Tactile Sensors for Estimating Geometry and Force (Sensors 2017, PMC)