Frustum

The image the sensor records through the chosen lens: a checkerboard chart, or the ground plane with the figures and scenery of the chosen scene, with the frame and any image circle outlined. Its readings are in the results.

Distortion Simulator

Brown–Conrady coefficients on the chosen projection, or a distortion-over-field table fitted to them.

Sensor

Sony IMX290
Application
Sensor size
mm
mm
µm

1920 × 1080 px · 2.1 MP · 6.39 mm diagonal · Rolling shutter · Color / NIR

Lens

4 mm
mm
Lens coverage

Distortion

Coefficients
Model

OpenCV order k₁, k₂, p₁, p₂, k₃ on the projection’s normalized coordinates.

Scene

Chart
Show
m
mm
Pattern

Field of view

Horizontal
72.8°
Vertical
43.5°
Diagonal
81.5°

Frame

Radial distortion at edge
−5.64 %
Radial distortion at corner
−7.25 %
Square at centre
138 px

Camera matrix

fx
1379.31 px
fy
1379.31 px
cx
960.00 px
cy
540.00 px
Image size
1920 × 1080 px

Square pixels, fx = fy = f / pitch, and the principal point at the frame centre, in pixels from the top-left corner.

How it works

Brown–Conrady model

Here we use the Brown–Conrady model, the same model as OpenCV's projectPoints. On normalized image coordinates (x, y) with s = x² + y², the radial factor is R = (1 + k₁s + k₂s² + k₃s³) / (1 + k₄s + k₅s² + k₆s³), and the point moves to x′ = x·R + 2p₁xy + p₂(s + 2x²), y′ = y·R + p₁(s + 2y²) + 2p₂xy.

Projections

The normalized coordinates come from the chosen projection: rectilinear (x = X / Z) gives OpenCV's pinhole model, and equidistant (ρ = θ) with radial terms k₁–k₄ gives OpenCV's fisheye model.

Camera matrix and recorded field

The camera matrix is the same as OpenCV's: square pixels with fx = fy = f / pitch and the principal point (cx, cy) at the frame centre, in pixels from the top-left corner. The recorded field of view undistorts the pixel at the middle of the right edge, the middle of the bottom edge and the corner, takes each one's field angle through the projection, and doubles it; barrel distortion packs more of the world into the frame than the ideal mapping, pincushion less.

Lens tables

With the Lens table model, the table becomes distortion relative to the projection, and k₁…kₙ are fitted to it by least squares. The image is drawn by inverting the mapping for every output pixel, and is checked against OpenCV projectPoints reference values.