マシンビジョンにおける迷光：ベールフレア、ゴースト、グレア、およびBBARコーティングによる対策
Veiling flare, ghosting, and target-side glare are different problems with different fixes. BBAR coatings, lens design, and filters address each one.
Stray light is non-image-forming light that reaches the sensor after reflecting or scattering inside the lens, not a bright object in the scene. It raises the dark-region floor, compressing contrast until low-luminance detail is unresolvable. Veiling flare spreads it as haze. Ghosting forms a localized secondary image. Target glare is a separate, scene-side problem. Fixing stray light means addressing the optics: lens design, coatings, window count, and shielding.
マシンビジョンにおける「迷光」の意味
Stray light is any light reaching the sensor that did not follow the lens's intended image-forming path: light reflecting off the inner barrel wall, scattering off a polished element edge, bouncing between two surfaces, or scattering from dust near the aperture stop.
The result is a background signal on top of the real image, which lifts the shadow floor and compresses usable dynamic range. It shows up as:
- ピントが合っている場合でも、MTF値と見かけのシャープネスが低下する
- シーンの輝度が高まるにつれて悪化するコントラストの低下
ISO 9358 quantifies this floor as the veiling glare index (VGI): the fraction of scene luminance reaching the sensor as non-image-forming light, against a defined target and black reference. A 1% VGI caps surviving contrast near 100:1, the 40 dB figure under a 20 log convention. A different target or convention moves that number.
多要素レンズに迷光が発生する理由
Every air-glass interface reflects a fraction of incident light. An uncoated surface reflects about 4-5%. A well-coated surface reduces that to 0.1-0.5%, but not to zero.
Ten surfaces give 45 surface pairs that can fold light back toward the image plane, plus higher-order paths and the sensor cover glass. One reflection alone mostly costs transmission; the second is what sends light forward again, and which pairs make a visible ghost depends on the stop, focus, and surface geometry. The design task is keeping those reflections off the sensor area, absorbing them in baffles (internal blackened rings that trap stray rays), or scrambling them into the noise floor rather than a coherent artifact.
迷光とグレア、フレア、ゴーストの違い
These terms get used interchangeably, but the fix depends on which problem is present. The table maps each failure mode to its usual cause and fix.
| 問題の種類 | 視覚症状 | 一般的な原因 | 考えられる解決策 |
|---|---|---|---|
| 迷光（一般） | コントラストの低下、シャドウの底上げ、フレーム内のどこかに生じるヘイズやアーティファクト | センサーに到達する光路内の、像を形成しない光 | レンズ設計（バッフル、AR/BBARコーティング）、窓の数の削減、シールド |
| ベールフレア | 画面全体に均一なかすみが生じており、暗い部分が黒ではなく灰色に見えてしまう | 複数の内部表面からの拡散散乱が、広いバックグラウンド信号として蓄積される | AR/BBARコーティングを施した低散乱レンズ、保護窓の数を削減、レンズフード |
| ゴーストフレア | 明るい光源からずれた、構造化された二次像（リング、円盤、筋） | 2つの表面間の反射により、センサー面上またはその付近に、通常は焦点が合っていない二次像が形成される | ゴーストの少ない光学設計。明るい光源を、ゴーストが発生しやすい視野角から遠ざける |
| ターゲットグレア | シーン内のオブジェクトに飽和した鏡面ハイライトが見られ、反射点でディテールが失われている | シーンの被写体からレンズへ直接入射する鏡面反射 | 交差偏波、照明配置、照射角の変化 |
A camera losing contrast outdoors could have any of these, and each needs a different fix: a polarizer will not reduce ghosting from internal reflections, and a lens hood will not remove a specular highlight from a shiny part.
マシンビジョンにおけるグレアをどのように軽減すればよいでしょうか？
Diagnose the failure mode first. Specular highlights on non-metallic surfaces call for cross-polarization; ambient contamination for a bandpass filter matched to the illumination wavelength; sensor clipping for less exposure or an ND filter.
非金属表面における交差偏光
Cross-polarization uses two linear polarizers 90 degrees apart: one over the source, one (the analyzer) over the lens. Specular reflection off a smooth non-metallic surface keeps its polarization and is blocked. Diffuse light depolarizes and partly passes, so surface detail survives. It works best on glossy dielectrics: plastics, glass, ceramics, solder mask. It can still reduce specular glare from bare metals, but less predictably, because metallic reflection alters the polarization state and the result depends on the metal, its surface roughness, and the incidence angle.
Each polarizer transmits about 50% of unpolarized light, so cross-polarization costs roughly two stops of signal. Set the F# from the depth of field the inspection needs, then add illumination to cover the loss, since lighting in most machine vision cells is programmatically controlled. Opening the iris on a Commonlands C-mount lens recovers that light only while the shallower depth of field stays acceptable.
バンドパスフィルターとNDフィルター
A bandpass filter passes a narrow band matched to the illumination source and blocks out-of-band ambient light, stabilizing contrast between day and night. It does not suppress specular reflection at the target wavelength. Neutral-density filters cut transmission uniformly but recover nothing lost to specular geometry. See the bandpass filter guide for selection by wavelength.
HDRと屋外シーンがなぜこの問題を浮き彫りにするのか
Indoors, with controlled lighting and a matte target, stray-light contributions from any single path are small, and a 0.1% floor is invisible. Outdoors, direct sunlight runs about 100,000 lux and deep shadow near 10 lux. Illuminance is not the ratio the sensor sees, since reflectance, BRDF, view angle, and atmospheric scatter intervene, but a scene lit across that range still hands the lens a luminance ratio in the thousands. Bright sources like sun, sky, and headlamps can appear anywhere, including just outside the field of view.
Take a luminance range of 10,000:1 and normalize the peak to 1.0, so shadow detail sits at 0.0001. A 1% veiling-glare floor adds 0.0100 everywhere, so that shadow signal now rides on a floor 100 times larger than itself, leaving about 1% local modulation, below what the pipeline and shot noise preserve.
視野外の軸外光源
A lens keeps admitting light past its specified field of view. Sources well outside it still enter the barrel and reach the sensor by reflection. Which angles hurt follows the barrel, its baffles, and element geometry, so take the danger band from an angular stray-light scan of the lens rather than a generic number.
A stray-light-optimized lens absorbs these in baffles. One that is not produces haze or ghosts that seem to come from nowhere. This matters most for fixed automotive and outdoor cameras that cannot avoid bright sources.
HDRセンサーがその性能を発揮するには、迷光の少ないレンズが必要です
HDR sensors reach 120 dB or more of dynamic range, but the lens sets the ceiling: a stray-light floor holding surviving contrast near 60 dB on the same convention throws most of that away. Commonlands treats a stray-light-optimized lens as close to a prerequisite for an HDR sensor outdoors.
BBARコーティングとは何ですか？また、どのようにして迷光を低減するのですか？
BBAR stands for broadband anti-reflective coating: a multilayer thin-film coating that reduces Fresnel reflection across a wavelength band rather than at a single design wavelength. A BBAR-coated surface typically reflects under 0.5% across the band, against roughly 4-5% for uncoated glass. Lower surface reflection means more light reaches the sensor and less bounces between elements.
A single quarter-wave AR layer gives a broad but shallow minimum (roughly 1.2-1.4% for MgF2 on crown glass). A V-coat drives reflection near zero at one wavelength, good for a single NIR line but poor elsewhere. BBAR instead holds low reflectance across a band, commonly 400-700nm for visible or 400-900nm for combined visible and NIR use.
BBAR is applied to individual elements during manufacturing, so a datasheet note like "BBAR on Lens 1 Surface 1" means exactly one surface in the stack carries it. The other surfaces may use standard AR, MgF2, or nothing.
BBAR is also distinct from a hydrophobic coating. BBAR reduces Fresnel reflection, while a hydrophobic coating is a fluoropolymer layer on the front element that repels water and oil. It still sits on top of the optical stack, so ask for measured reflectance over wavelength and angle with the overcoat applied instead of assuming it is negligible. A lens can specify both as separate features.
| コーティングの種類 | 何が減少するのか | 解決できないこと |
|---|---|---|
| BBAR | コーティング面における波長帯にわたるフレネル反射 | コーティングされていない表面、レンズ筒の壁面、またはバッフルの隙間からの迷光；歪み；収差 |
| 狭帯域AR | 特定の波長（例：850nm）における反射 | 他の波長域における反射；可視光と近赤外光の併用 |
| 疎水性コーティング | 前面への水、油、指紋の付着 | Ghosting, flare, and throughput set by the rest of the stack; it is not an AR design |
Verify coating claims against the current datasheet: the design band, which surfaces carry the coating, and average reflectance (Ravg) across the band all vary by product. With BBAR on only one or two surfaces, ghosting and flare still depend on element count, baffling, and barrel design.
One failure to watch: if a coating covers only 400-700nm but the system runs active NIR, the uncoated NIR reflections create ghosts that are invisible at the bench but clear in sensor output. Match the coating band to every illumination source.
エンジニアが実際にどのように迷光を低減しているか
There is no single universal fix. The approach depends on which part of the optical system generates the stray light.
The first tool is anti-reflection coating on every surface, since every uncoated interface reflects several percent. A stray-light-optimized design adds black-coated barrel baffles, matte aperture stops, and attention to which surfaces can form second-order reflections toward the sensor.
Every extra surface, including a flat protective window, adds to the budget. For environmental protection, an IP-rated lens that seals internally usually costs less stray light than a separate window. A sealed M12 lens such as the Commonlands CIL034 in its M12A variant, rated IP67, provides outdoor protection with no additional air-glass interfaces. The M12B build of the same optics carries no ingress rating, so check the variant suffix before assuming a lens is sealed.
A window still earns its surfaces where impact, abrasion, chemical exposure, field replacement, or sealing a whole enclosure is the requirement.
A lens hood or housing extension blocks off-axis sources before they reach the front element, one of the lowest-cost fixes for fixed-mount cameras. Where packaging limits hood size, as in automotive, the lens design has to do more. With programmable lighting, arranging sources out of the direct lens view lowers stray-light loading.
Image processing is a fallback, not a substitute. Flat-field correction addresses fixed multiplicative shading and pixel-gain non-uniformity, not the additive, scene-dependent background that veiling flare adds, so it does not generally remove flare; only a background that stays fixed can be subtracted by a separate calibration, and neither approach removes ghosts that move with the source.
迷光対策が施されたM12レンズおよびグレア抑制用アクセサリー
Commonlands stocks optical filters for ambient rejection below. A sealed IP-rated M12 lens can replace a separate protective window in outdoor builds.
迷光およびグレアのトラブルシューティングチェックリスト
Work through these when an outdoor, HDR, or reflective-scene camera shows unexpected contrast loss, haze, or ghost artifacts.
- Identify the artifact. Uniform haze (veiling flare), a structured spot displaced from a bright source (ghosting), or a saturated highlight on one surface (target glare). Each maps to a different fix.
- Capture a frame with the illuminator off. Usable signal from ambient alone points to a bandpass filter matched to the illumination wavelength.
- Move the light or camera 10-20 degrees. A highlight that tracks the source is specular geometry, not internal stray light.
- Count optical surfaces in the path. An uncoated window adds two 4-5% reflection surfaces. Where the window is only keeping water out, an IP-rated lens that seals internally does the same job without adding any.
- Test cross-polarization on non-metallic parts, and expect to add illumination or open the aperture to compensate. Reduced exposure that recovers gradient means the highlight was clipping, not a geometry issue.
- Confirm the lens is validated as low-ghost or stray-light-optimized for HDR or outdoor use, then verify with a hood on and off before locking the build.
よくある質問
Commonlands lenses use anti-reflection coatings to cut internal reflections. These answers explain what coatings can and cannot fix.
マシンビジョンシステムにおける「迷光」とは何ですか？
迷光とは、レンズを通る本来の撮像経路をたどらずにセンサーに到達する光のことです。これは、レンズ内部の筒面での散乱、レンズ要素の縁での反射、あるいは光がまっすぐに通過せずに表面間で跳ね返ることによって生じます。単に被写体内の明るい物体によるものではありません。これはすでに光路内にある不要な光であり、その影響として背景信号が生じ、コントラストを低下させ、低輝度域では被写体の細部が判別できなくなることがあります。
グレア、フレア、ゴースト、迷光の違いは何ですか？
迷光とは、光学系内部で像を形成しない光がセンサーに到達する現象の総称です。ベールフレアは、その光を均一な霞のように広範囲に拡散させ、明確な構造を伴わずにコントラストを低下させます。ゴーストは局所的なアーティファクトであり、特定の面と面の間の反射によって生じる構造化された二次像です。グレアはこれとは異なり、被写体自体からの飽和した鏡面反射であり、被写体側の問題です。 偏光フィルターや光学系の配置変更で被写体のグレアを解消しても、内部の迷光は解消されず、その逆も同様です。
マシンビジョンにおけるグレアをどのように軽減すればよいでしょうか？
まず、不具合の原因を特定してください。非金属表面の鏡面反射によるハイライトについては、交差偏光法を使用します。具体的には、照明装置に直線偏光フィルターを装着し、レンズ側に交差した分析フィルターを設置します。周囲の光による干渉については、照明の波長に合わせたバンドパスフィルターを使用してください。センサーのクリッピングについては、露光時間を短縮するか、NDフィルターを追加してください。照明の配置を変更して、鏡面反射の角度を軸から外すことは、多くの場合、最も手っ取り早い最初のテストとなります。
レンズのBBARコーティングとは何ですか？
BBARとは「ブロードバンド反射防止コーティング（Broadband Anti-Reflective Coating）」の略称であり、特定の設計波長ではなく、広範な波長帯にわたってフレネル反射を低減する多層薄膜コーティングです。コーティングを施していないガラスは、表面ごとに入射光の約4～5％を反射しますが、BBARコーティングを施した表面では、通常、設計波長帯全体で反射率が0.5％未満に抑えられ、透過率が向上するとともに、フレアやゴーストの原因となる内部反射が低減されます。
保護用ウィンドウは、どのような場合に迷光を悪化させるのでしょうか？
A protective window adds at least two new air-glass interfaces. Uncoated, each reflects roughly 4-5% of incident light; even AR-coated, a few tenths of a percent remain. Those accumulate, so a standard flat window in front of a well-designed lens can increase veiling flare in high-dynamic-range scenes. An IP-rated lens that seals at the front element keeps those interfaces out of the budget, though a window still wins where impact, chemical exposure, or field replacement is the requirement.
ご用途に適した迷光の少ないレンズの選定でお困りではありませんか？
Commonlands engineering can help identify whether stray light, ghosting, or target glare is the root cause in your build, then match the lens, coating, or filter to your sensor format and environment.