We know that when rotational and linear velocity of the camera is high it is highly likely to get blurred images. The blurred image affects the quality of the accuracy in the feature tracking problem. Assuming that these velocities are known, how do we penalize the period of the low quality tracks?

For example, let $\textbf{p}_i\in R^3$ be $i_{th}$ 3d point and $\textbf{z}_{ij} \in R^2$ $j_{th}$ projected feature at $i_{th}$ camera pose.

Than, the objective function in the bundle adjustment problem is defined as

$\textbf{e} = \textbf{z}_{ij} - \pi(\textbf{T}_j\textbf{p}_i)$

If we know the velocity at $\textbf{T}_j$ how do we apply this information to the optimization?

An instant idea came to my mind is utilizing the velocity in terms of

$f = \textbf{e}^T\Sigma_j^{-1}\textbf{e}$

where the covariance is $\Sigma_j=I_{3\times3}v$. v is velocity. It might work but looks huristic.

Anyone knows a related literature?


Your Answer

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

Browse other questions tagged or ask your own question.