The following equations are Kruppa's simplified equations used in camera autocalibration. My objective here is to infer the cost function(Error Function) from this equations, So I can minimize the cost function through Levenberg-Marquardt nonlinear optimization algorithm.

(I already did the part of optimization, and only need the cost function in mathematical form so I can implement it).

I have read this paper, and this paper about this specific topic, but I didn't understand the complicated math inside.

Can anybody explain how the mathematical cost function will look like in simple language?

Thanks in Advance

Kruppa $ W^{-1} = K K^{T}$,

where $W^{-1}$ is the dual of $IAC$, $K$ is the camera intrinsics matrix

$r,s$ are the first, and the second singular values of $S$

$u_1, u_2, u_3$ are the column-vectors of $U$,

$v_1, v_2, v_3$ are the column-vectors of $V$,

$USV = SVD(Fundemental\quad Matrix\__{ij})$,

and $i,j$ are two different images from moving camera.


According to the equation(30) in this paper The solution was to reform the Equations from the picture in my question here, and after that(Which the paper mentioned in different, complicated mathematical format) is to take the Frobenius norm to the subtract of newly formulated equations(30) on the paper.

$\rho_{1}/ \phi_{1} = \rho_{2}/ \phi_{2} = \rho_{2}/ \phi_{2}$ where $\rho$ is the numerator of each term of the equations, and $\phi$ is the denominator of each term, now the equation(30) in the paper proposed that: $\rho_{1}* \phi_{2} - \phi_{1}* \rho_{2} = 0 $ and $\rho_{1}* \phi_{3} - \phi_{1}* \rho_{3} = 0 $ now taking the Frobenius norm $\parallel.\parallel_{fro} $ for the two equations yields two values, Thus the cost function is the sum of this values.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.