Multi-Direction Sequential Imaging for Surface Normal Reconstruction
- 4+ LED illuminators at known azimuths activated sequentially capture brightness response per illumination direction.
- Lambert cosine law inversion per pixel solves for the 3-component surface normal across the entire image.
- Decouples albedo from topography — produces separate 2D reflectance map and 3D shape map per inspection cycle.
- Best fit for embossings, dents, scratches, weave structures, low-contrast DPM requiring 3D feature separation.
- Elevation 30–60° balances specular interference and self-shadowing on industrial surfaces.
- Polarised variants required on glossy targets to suppress specular component and preserve Lambert assumption.
Photometric stereo is a 3D imaging technique that reconstructs surface normals and decouples topography from albedo by acquiring multiple images of the same target under different illumination directions. By exploiting the dependence of image brightness on the angle between the local surface normal and the illumination direction, the technique extracts the three-dimensional orientation of every pixel without requiring a calibrated geometric setup like structured light. Photometric stereo is particularly powerful for inspection of textured surfaces, embossed features and subtle three-dimensional defects.
Working Principle of Photometric Stereo
A photometric stereo system consists of four or more LED illuminators positioned around the camera at known angles, typically at 90 degrees apart in azimuth and at a fixed elevation above the inspection surface. The camera is fixed and observes the target perpendicular to its plane. The illuminators are activated in sequence, with each illuminator capturing a separate image under directional illumination from one specific azimuth. Industrial-grade photometric stereo arrays are engineered within the Custom LED Illuminators portfolio, often paired with synchronised multi-azimuthal LED Ring Illuminators for compact deployment.
Each pixel in each image has a brightness that depends on the local surface normal and the illumination direction, following Lambert’s cosine law for diffuse surfaces. By combining the four (or more) brightness values at each pixel with the known illumination directions, the algorithm solves for the three components of the local surface normal at each point. The resulting normal map represents the local surface orientation across the entire image and can be integrated to produce a height map or processed directly to highlight three-dimensional features.
Decoupling Albedo from Topography
A unique advantage of photometric stereo is its ability to separate the local reflectance (albedo) of the surface from its three-dimensional shape (topography). The algorithm produces two independent images at each pixel: an albedo image that shows the intrinsic reflectance variations of the surface (printed marks, colour patterns, surface staining) and a topography image that shows only the three-dimensional features (embossings, scratches, dents). This decoupling is impossible in standard 2D imaging and is the primary motivation for adopting photometric stereo in many applications.
Typical Industrial Applications
Photometric stereo is essential for inspection of embossed features on packaging, tablets and structural plates; detection of subtle dents, scratches and surface deformations on automotive body panels; quality control of textile surfaces where weave structure must be analysed separately from dye colour; inspection of leather products for embossing and surface defects; quality control of moulded plastic parts for surface flow lines and three-dimensional defects; verification of printed paper and cardboard for embossed elements; reading of low-contrast embossed codes on metal and plastic components; and any application where the three-dimensional aspect of the surface must be separated from its two-dimensional pattern.
Selection Criteria and Design Considerations
The number and arrangement of illuminators determine the accuracy and robustness of the surface normal estimation. Four illuminators at 90-degree azimuthal spacing represent the minimum configuration and provide adequate performance for general applications. Eight illuminators at 45-degree spacing provide better robustness to shadows and saturation effects on complex surfaces. Higher numbers approach a continuous angular distribution and produce the highest accuracy at the cost of longer acquisition cycles.
The elevation angle of the illuminators is the second design parameter. Steeper elevation (close to vertical) produces strong specular reflections on glossy surfaces that interfere with the Lambert assumption. Lower elevation (close to grazing) produces strong shadows that mask portions of the image. Optimal elevation angles are typically between 30 and 60 degrees from the surface plane.
Synchronisation and Acquisition Time
Each photometric stereo cycle requires N sequential images (where N is the number of illuminators), which multiplies the total acquisition time by N relative to a single-shot imaging system. For inspection at high speeds, the LED switching and camera triggering must be precisely synchronised to ensure that each image corresponds to exactly one active illuminator. The required multi-channel synchronisation is delivered by the RODER LED drivers and electronic controllers catalogue.
Integration and Limitations
Photometric stereo systems integrate as multi-illuminator fixtures with synchronised drivers, typically packaged as a single unit that includes the LED array and the camera trigger logic. The output of the system can be the raw image stack (left to the application software for analysis) or pre-processed albedo, topography and normal maps.
The principal limitation of photometric stereo is its dependence on the Lambert assumption, which holds well for matte and semi-matte surfaces but breaks down for highly specular targets. On glossy surfaces, the bright specular reflections introduce errors in the normal estimation and can compromise the reconstruction. Polarisation can be used to suppress the specular component and recover the diffuse signal, at the cost of brightness loss.
The second limitation is the longer acquisition time compared to single-shot imaging, which limits the maximum production speed. For high-speed lines, multi-camera configurations with simultaneous capture under different illuminations may be required, increasing the cost and complexity of the system.
RODER Vision Photometric Stereo LED Arrays
RODER Vision engineers application-specific photometric stereo LED arrays with synchronised multi-channel drivers for industrial vision inspection requiring surface normal reconstruction and albedo/topography decoupling.
- Application-specific 4, 8 or 16-direction photometric stereo arrays — Custom LED Illuminators
- Multi-azimuthal ring configurations for compact photometric stereo deployment — LED Ring Illuminators
- Multi-channel synchronised drivers with microsecond switching for sequential acquisition — LED Drivers and Electronic Controllers
Photometric stereo installations require shielded multi-channel cabling — the RODER catalogue includes industrial-grade cables and fastening systems engineered for synchronised multi-illuminator deployments.