Camera Hardware: Optics → Sensor → Wire

Stage 1 · Optics — shaping the photons

The lens decides three things before the sensor ever sees light: field of view, depth of field, and how much light gets through. The classic equation:

• Focal length ≈ zoom. 16mm = wide, 35mm = portrait, 200mm = sports/wildlife.
• f-stop = focal length ÷ aperture diameter. Smaller f-stop = wider opening = more light + shallower DoF.
• Depth of field shrinks with longer focal length, larger aperture, and shorter focus distance.

Visualizer · lens geometry & depth of field

Lens artifacts — what real glass does to your image

The pinhole model is a lie. Real lenses introduce:

• Radial distortion — barrel (wide-angle) or pincushion (tele). Straight lines in the world become curves on the sensor. Modeled by polynomial coefficients k₁, k₂, k₃ applied to r².
• Chromatic aberration — different wavelengths refract at different angles, so red/blue channels misalign at high-contrast edges.
• Vignetting — light falloff toward the corners (the cos⁴(θ) law plus mechanical limits).
• Diffraction — at very small apertures (f/16+), the wave nature of light caps your sharpness regardless of focus.

Calibration software (OpenCV, Kalibr) recovers all of this from a checkerboard sequence and produces an undistortion map you apply per pixel.

Visualizer · radial distortion & CA

Drag the distortion slider to barrel- or pincushion-warp a checkerboard. Add chromatic aberration to see how RGB channels separate at the edges.

Stage 2 · Exposure — three knobs, one EV

Once you've chosen a lens, three controls decide image brightness:

• Aperture — already set by your f-stop, also affects DoF.
• Shutter speed — how long the sensor integrates. Longer = brighter + motion blur.
• ISO / Gain — sensor multiplier. Higher ISO = brighter + more noise.

Visualizer · exposure triangle

Stage 3 · Sensor — photons to bits

Photons hit silicon → photoelectrons (charge) → voltage → ADC integer → pixel.

• Well capacity — max electrons a pixel can hold before saturating. Bigger pixels = bigger wells = better dynamic range.
• Bit depth — 8-bit = 256 levels, 12-bit = 4096 levels. More bits = more headroom under low light, fewer banding artifacts.
• Read noise — electrons of uncertainty added by the readout circuit. Low ISO + cooled sensors minimize it.

Visualizer · electrons to pixel

The Bayer mosaic

Color sensors put a Bayer filter over the pixels: R, G, G, B in a 2×2 tile (RGGB). Each pixel sees only one color. The other two are reconstructed from neighbors via demosaicing. Twice as many green pixels because human vision is most sensitive to green — the same reason the grayscale formula is 0.299R + 0.587G + 0.114B.

Monochrome sensors drop the Bayer filter entirely. Every pixel sees every photon. Result: better SNR, sharper edges, no demosaic artifacts. For machine-vision inspection where color isn't needed, mono is almost always the right call.

Visualizer · Bayer mosaic & demosaic

Stage 4 · The wire — bandwidth math

A handful of reference points:

• 1 MP mono, 30 fps, 8-bit → 30 MB/s = 240 Mb/s. Comfortable on USB 3.0.
• 4K color, 60 fps, 10-bit → ~17 Gb/s raw. Needs 10GbE + compression, or CoaXPress.
• 20 cameras × 2 MP × 30 fps on a robot → 7.2 Gb/s combined. PoE switch + GigE Vision + JPEG.

Protocols at a glance: USB 3.0 Vision (5 Gb/s, 3 m), GigE Vision (1 Gb/s, 100 m), 10 GigE, CoaXPress (6.25 Gb/s on coax, 100 m+), MIPI CSI-2 (on-board only).

Visualizer · link budget