Technique d’éclairage direct en computer vision

In the field of industrial automation, if backlighting is the tool for shape, direct illumination is the tool for detail. While backlighting creates a binary world of silhouettes, direct illumination (episcopic lighting) opens the door to the complex reality of surface textures, colors, topography, and material properties.

For the vision engineer, direct illumination is the art of controlling how light interacts with a surface to create high-quality contrast. Whether you are identifying a microscopic scratch on a polished metal plate or reading a laser-etched DataMatrix code on a curved plastic component, the geometry and quality of your direct lighting setup will determine the robustness of your deployment.

The Physics of Surface Interaction

Direct illumination relies on the light reflecting off the object and into the camera lens. To master this, one must understand the two primary types of reflection dictated by the surface’s micro-geometry.

Specular vs. Diffuse Reflection

When a light ray hits a surface, its behavior is governed by the Law of Reflection.

$$\theta_i = \theta_r$$

• Specular Reflection: Occurs on « mirror-like » surfaces (polished metal, glass, liquids). The light reflects at a single outgoing angle. In computer vision, this often leads to « hot spots » or glare that can blind the sensor.
• Diffuse Reflection: Occurs on rough or matte surfaces (paper, unpolished wood, cast iron). The light is scattered in many directions (Lambertian reflection). This is generally easier to image because the object appears consistently bright from multiple viewing angles.

The Inverse Square Law and Cosine Law

The intensity of light reaching the object ($E$) is inversely proportional to the square of the distance ($r$) from the source and proportional to the cosine of the angle of incidence ($\theta$):

$$E = \frac{I}{r^2} \cos \theta$$

For a vision professional, this means that even minor changes in the mounting distance or the tilt of a part can lead to significant fluctuations in image brightness, potentially breaking threshold-based inspection algorithms.

Taxonomy of Direct Illumination Strategies

Different surface challenges require different lighting geometries. We can categorize direct illumination into four primary technical strategies.

A. Brightfield Illumination (Partial)

Brightfield is the most common technique. The light source is positioned at a relatively high angle relative to the surface (typically $45^{\circ}$ to $90^{\circ}$).

• Mechanism: Light reflects off the surface directly into the lens.
• Result: Flat, reflective surfaces appear bright; features that scatter light (like scratches or pits) appear dark.
• Best for: General presence detection, color inspection, and non-reflective surfaces.

B. Darkfield Illumination

Darkfield lighting involves placing the light source at a very low angle (typically $10^{\circ}$ to $30^{\circ}$) relative to the object surface.

• Mechanism: Most light reflects off the flat surface and away from the lens. Only light that hits a raised or recessed feature (topography) is scattered into the camera.
• Result: The background remains dark, while surface defects, dust, and textures « glow » brightly.
• Best for: Scratch detection on polished surfaces, inspecting embossed text, and identifying contaminants on flat substrates.

C. Coaxial (On-Axis) Illumination

Coaxial lighting uses a semi-transparent beam splitter to align the light path perfectly with the optical axis of the lens.

• Mechanism: Light is reflected onto the object from the same direction the camera is looking.
• Result: It eliminates shadows and is particularly effective for flat, specular surfaces.
• Best for: Semiconductor wafer inspection, PCB fiducial alignment, and reading codes on highly reflective packaging.

D. Diffuse (Dome/Cloudy Day) Illumination

Reflective, curved, or irregular surfaces (like a crushed soda can or a blister pack) create « specular glare » that ruins images. Diffuse illumination uses a dome or a large scattering surface to provide light from all angles ($180^{\circ}$ hemisphere).

• Mechanism: It mimics a « cloudy day » where light comes from everywhere, eliminating directional shadows and hot spots.
• Best for: Highly reflective curved surfaces, metal cylinders, and inspecting objects through transparent but wrinkled overwraps.

Comparison of Lighting Geometries

Technique	Setup Complexity	Best Surface Type	Objectif principal
Illuminateur annulaire	Faible	Matte / Diffuse	General feature recognition
Darkfield	Medium	Polished / Mirror	Surface defect detection
Coaxial	Élevé	Flat / Reflective	Eliminating shadows on shiny parts
Dome Light	Élevé	Curved / Specular	Eliminating glare on complex shapes

Choix de la longueur d’onde et filtrage optique

The color of the light determines how it interacts with the material’s molecular structure.

The Color Wheel Principle

In direct illumination, we use the color wheel to enhance or suppress features:

• To make a feature look bright: Use light of the same color (e.g., Red light on a red part).
• To make a feature look dark: Use light of the opposite color (e.g., Green light on a red part).

Infrared (IR) vs. Ultraviolet (UV)

• Infrared ($850\text{–}940\text{ nm}$): Excellent for reducing « visual noise. » IR can penetrate certain inks and plastics, making it ideal for checking fill levels through colored plastic or ignoring distracting print on a package.
• Ultraviolet ($365\text{–}395\text{ nm}$): Used primarily for Fluorescence Inspection. Many adhesives, oils, and specialized inks glow under UV, allowing the vision system to detect their presence on surfaces where they would otherwise be invisible.

Polarization: The Glare Killer

When dealing with extreme specular reflection (like « hot spots » on a wet surface), Polarization is an essential tool. By placing a linear polarizer on the light source and another (an analyzer) on the lens oriented at $90^{\circ}$, the vision system can « extinguish » the direct specular reflection while allowing the depolarized scattered light through. This significantly increases the dynamic range of the image.

Integration and Control

Strobe and Pulse Management

For high-speed lines, continuous lighting often provides insufficient intensity for short exposure times, leading to motion blur. Professional systems utilize Strobe Controllers to synchronize the light pulse with the sensor’s Global Shutter.

• Overdriving: Pulsing an LED at $10\text{x}$ its rated current for a few microseconds can provide the intensity needed to « freeze » a part moving at several meters per second.

Mounting and Stability

In direct illumination, the angle is everything. A shift of $1^{\circ}$ in a darkfield setup can be the difference between seeing a defect and seeing nothing. Industrial mounting solutions must be vibration-resistant and provide 3-axis adjustment to maintain the « sweet spot » of the optical path.

Advanced Software Considerations

While direct illumination aims to provide the best raw image, software algorithms must still handle the inherent variability of incident light.

1. Gain Control vs. Exposure: Always prioritize increasing light intensity or exposure time over digital Gain. Gain amplifies the noise along with the signal, reducing the precision of edge-detection algorithms.
2. Flat-Field Correction (FFC): No light source is perfectly uniform. FFC is a software calibration technique that compensates for the « vignetting » or unevenness of the light source, ensuring that a pixel in the corner has the same response as a pixel in the center.
3. High Dynamic Range (HDR): In cases where a single image cannot capture both dark details and bright highlights (common in direct lighting), multiple exposures can be combined to create a high-contrast master image.

Conclusion

Direct illumination is the most versatile—yet challenging—component of a machine vision system. Success requires a deep understanding of the interplay between light geometry, material properties, and sensor physics. By selecting the correct strategy—whether it be the subtle topography revealed by darkfield or the glare-free uniformity of a dome—engineers can ensure that their computer vision algorithms are fed with the highest quality data possible.

The goal of professional lighting is simple but profound: make the « good » parts look obvious and the « bad » parts look impossible to miss.

---

Caractéristique	Rétroéclairage	Éclairage direct	Éclairage diffus
Direction de la lumière	Derrière l’objet	Avant/côté de l’objet	Omnidirectionnel / diffus
Objectif principal	Silhouette & Géométrie	Détails de surface & Couleur	Suppression des reflets et ombres
Idéal pour	Métrologie, inspection de trous	Codes-barres, texture, rayures	Pièces brillantes, courbées ou métalliques
Avantage clé	Contraste maximal des bords	Extraction de détails fins	Élimine les « hot spots » (reflets)
Défi principal	Nécessite un accès des 2 côtés	Gestion des reflets spéculaires	Encombrement physique (dôme) & intensité
Optique idéale	Optiques télécentriques	Optiques standard / macro	Optiques standard / grand angle