Calibrating Conic in Camera Calibration

• The calibrating conic is a real conic in the image plane that encodes intrinsic camera parameters through both algebraic and geometric properties.
• It establishes a closed-form relationship between the camera calibration matrix and observable features like vanishing points and fixed-angle rays.
• It underpins practical applications such as angle measurement, self-odometry, and perspective recovery in computer vision and metrology.

A calibrating conic, sometimes called the calibrated conic or image of the calibration conic, is a central real conic in the image plane of a projective camera encoding intrinsic geometric structure. Born from classical constructions in projective geometry and now essential in computer vision, metrology, and geometric analysis, the calibrating conic provides a direct visual and algebraic locus relating intrinsic camera parameters to observable image features, such as vanishing points and rays of fixed angle relative to the optical axis. Its role is both representational—apparent in conic matrix algebra—and operational, underpinning angle measurement, intrinsic calibration algorithms, and geometric invariants. This article surveys the principles, explicit algebraic forms, computational algorithms, and major applications of the calibrating conic and its geometric companion, the conformal point, drawing on foundational and recent research (Hartley, 16 Jan 2026).

1. Algebraic and Geometric Definition

Let $K$ be the $3\times 3$ camera calibration matrix, parameterized in pixel units as

$$K = \begin{pmatrix} \alpha & s & u_0 \ 0 & \beta & v_0 \ 0 & 0 & 1 \end{pmatrix}$$

where $\alpha,\beta$ are the focal lengths along $x$ and $y$, $s$ the skew, and $(u_0,v_0)$ the principal point. The calibrating conic $C$ is defined in homogeneous image coordinates $x=(x, y, 1)^\top$ by

$$C \sim K^{-T} D K^{-1}, \qquad D = \mathrm{diag}(1,1,-1).$$

Equivalently, $C$ is the locus in the image plane of points that correspond to rays forming a $45^\circ$ angle with the optical axis: $$x^\top C x = 0 \ \Longleftrightarrow\ \|K^{-1}x\|^2 = (K^{-1}x)_3^2.$$ Unlike the image of the absolute conic (IAC) $\omega = (K K^\top)^{-1}$, which is imaginary and thus not directly visualizable, $C$ is a real central conic—typically an ellipse in photographic cameras—centered at the principal point $(u_0,v_0)$ (Hartley, 16 Jan 2026).

2. Relationship to Camera Calibration and Intrinsic Parameters

The calibrating conic $C$ encodes the full set of camera intrinsics up to scale. Its matrix, when written with $C_{33}=1$, has explicit closed-form entries: $$\begin{aligned} c_{13} & = -u_0, & \qquad c_{23} &= -v_0, \ c_{11} & = \frac{u_0^2 - f_x^2}{f_x f_y}, \qquad & c_{22} &= \frac{v_0^2 - f_y^2}{f_x f_y}, \ c_{12} & = \frac{ -u_0 v_0 + s f_y }{f_x f_y } \end{aligned}$$ where $f_x=\alpha$, $f_y=\beta$, $s$ is the skew, and $(u_0,v_0)$ is the principal point. Given $C$, the calibration matrix $K$ can be recovered in closed form by inverting $C$ or by direct algebraic manipulation: $$u_0 = -M_{13}, \quad v_0 = -M_{23}, \quad \alpha^2 = u_0^2 + M_{11},\quad \beta^2 = v_0^2 + M_{22}, \quad s = \frac{M_{12} + u_0 v_0}{\beta}$$ where $M = C^{-1}$, normalized so that $M_{33} = -1$ (Hartley, 16 Jan 2026).

3. Geometric Interpretations and Conformal Point

The conformal point $C_\ell$ is a unique image point associated to a chosen reference line $\ell$ (such as the image of the horizon) with the property that, for any points $A,B$ on $\ell$, the world angle between rays from the optical center through $A$ and $B$ is mapped to the planar angle $\angle A C_\ell B$ measured in the image. In the square-pixel, zero-skew case, if the principal point is $P=(u_0,v_0)$ and the reference line is a distance $d$ from $P$, $C_\ell$ lies on the normal from $P$ to $\ell$ at distance $\sqrt{f^2+d^2}$.

The calibrating conic $C$ is also the locus of tangent lines corresponding to rays leaving the principal point at $45^\circ$ inclination. It enables direct geometric constructions: e.g., reflected polars of points with respect to $C$ correspond to orthogonality conditions of vanishing points, and the center and axes of $C$ correspond visually to the principal point and focal lengths. The conformal point further enables ruler-and-compass constructions of angles, supporting visual computation of world angles between rays in the image (Hartley, 16 Jan 2026).

4. Computational Determination and Fitting

There are three principal classical methods for determining $C$ from image measurements:

• Three Orthogonal Vanishing Points: If three vanishing points in the image correspond to mutually orthogonal world directions, they must lie on $C$ and their triangle’s orthocenter yields the principal point. One fits the unique real conic passing through these points using the homogeneous quadratic equation.
• Two Orthogonal Directions plus Principal Point: Assuming square pixels and known principal point, two orthogonal vanishing points suffice; the reflected conic-polar construction of Hartley-Zisserman provides a closed-form solution.
• Known Angle Between Two Rays: Given image points $A$ and $B$ whose world rays make a known angle $\theta$, $C$ satisfies $$A^\top C B = \pm\cos\theta\sqrt{A^\top C A\,B^\top C B}$$; in this setup, planar geometry (e.g., the circle through $A,B$ subtending angle $\theta$ from the conformal point) can also be used to solve for $C$.

After estimating $C$, $K$ is extracted via the algebraic relationships detailed above. Even with noisy vanishing points, robust least-squares/SVD fitting plus these algebraic steps yield very accurate camera calibration (Hartley, 16 Jan 2026).

5. Applications in Geometric Computation and Vision

• Visual Angle and Direction Measurement: $C$ enables measurement of angles between image rays by algebraic or geometric means, avoiding direct reliance on $K$ or $\omega$.
• Intrinsic Parameter Visualization: The central position, axis lengths, and orientation of $C$ in the image directly reveal $(u_0, v_0, \alpha, \beta, s)$, making $C$ a visual embodiment of intrinsic calibration.
• Self-Odometry and Motion Estimation: Because the conformal point and $C$ remain fixed under pure planar motion, relative pose or rotation can be deduced by tracking image points and their angular relationships.
• Field of View and Scene Geometry: The axes of $C$ correspond to $45^\circ$ angular limits—its geometry encodes the camera’s field of view.
• Artistic and Architectural Analysis: In geometric analysis of paintings or drawings, fitting $C$ recovers the artist’s or viewer’s perspective and relative orientation.
• Robustness to Noise and Ruler-and-Compass Constructions: Unlike matrix-centric approaches, $C$ allows for geometric constructions solely in the image plane, favoring interpretability and tolerance to measurement uncertainty (Hartley, 16 Jan 2026).

6. Summary Table: Calibrating Conic Properties

Property	Algebraic Expression	Geometric or Computational Role
Conic matrix $C$	$K^{-T} D K^{-1}$	Encodes camera intrinsics; real central conic in image
Conformal point $C_\ell$	Construction via $C$ and reference	Unique viewpoint for angle preservation on reference line
Recover $K$ from $C$	Closed-form from $C^{-1}$	Full intrinsic extraction, usable for practical calibration
Fitting inputs	Vanishing pts, known angles	Image-feature-based calibration pipelines
Visualization	Drawn conic in image	Reads principal point, focal lengths, pixel skew from geometry

The calibrating conic is the unique real conic in the image plane whose algebraic and geometric properties fully encode and visually represent the camera’s intrinsic parameters. It enables both precise numerical calibration and intuitive geometric reasoning, providing a bridge between projective algebra, visual measurement, and practical camera calibration (Hartley, 16 Jan 2026).