BA optimization camera 3 d pose-CodePudding

Currently I only have a backlog of image, image feature points in the known world coordinates, with just three points of these feature points, the use of P3P algorithm for pose, so if I were to use the image of multiple points to three groups, the settlement to get multiple R, t matrix, can use the attitude matrix of the BA to multiple optimization?

CodePudding user response:

BA is doing this,

Do you know the true value and pixel coordinates, but due to the deviation exists, so the whole transformation matrix T is not the only,
Then define the loss function better, make optimization with BA,
When the loss can converge to the minimum value of T is the optimal solution,

Specific have to look at some examples to imitate, Pittsburgh and Google cesis or what, there are sample code

CodePudding user response:

reference 1st floor alittlenewbiek response:

BA is doing this,

Do you know the true value and pixel coordinates, but due to the deviation exists, so the whole transformation matrix T is not the only,
Then define the loss function better, make optimization with BA,
When the loss can converge to the minimum value of T is the optimal solution,

Specific have to look at some examples to imitate, Pittsburgh and Google cesis or what, have sample code

But I read on the Internet are mostly, BA is the optimization for multiple frames, but I'm the only one frame, just use a frame of image matching to pose more than three groups, with multiple R, t matrix, and then the R, t matrix optimization, it also ok? Error will be big?

CodePudding user response:

Because you have the world coordinates of a single image can also be, as long as there are multiple sets of points,

Didn't know it, but in general the world coordinates need more frames to write using more depending on the geometry of the relationship between the expressions of the loss,

A picture of the optimization can only guarantee that the effect of a transformation matrix is good,
But if your world coordinate and pixel coordinate itself must enough, also can be applied to the other picture transformation matrix,

As long as you can find the relation between, BA can application

P. s. written a book, the author zhang SLAM
And I can see multiple view geometry in CV

CodePudding user response:

reference 4 floor alittlenewbiek response:

because you have the world coordinates of a single image can also be, as long as there are multiple sets of points,

Didn't know it, but in general the world coordinates need more frames to write using more depending on the geometry of the relationship between the expressions of the loss,

A picture of the optimization can only guarantee that the effect of a transformation matrix is good,
But if your world coordinate and pixel coordinate itself must enough, also can be applied to the other picture transformation matrix,

As long as you can find the relation between, BA can application

P. s. written a book, the author zhang SLAM
And I can see multiple view geometry in CV

Excuse me, is only a frame diagram, use multiple R the P3P, t matrix using BA optimization and calculation result directly use EPNP what's the difference? The error between the two have a gap?

CodePudding user response:

reference 4 floor alittlenewbiek response:

because you have the world coordinates of a single image can also be, as long as there are multiple sets of points,

Didn't know it, but in general the world coordinates need more frames to write using more depending on the geometry of the relationship between the expressions of the loss,

A picture of the optimization can only guarantee that the effect of a transformation matrix is good,
But if your world coordinate and pixel coordinate itself must enough, also can be applied to the other picture transformation matrix,

As long as you can find the relation between, BA can application

P. s. written a book, the author zhang SLAM
And I can see multiple view geometry in CV

Eldest brother, you said zhang SLAM, didn't see ah, see only a xi 'an electronic science and technology, he is not home page what SLAM works

CodePudding user response:

This I don't know now, haven't see for a period of time,

The need to you from other sources to understand, because it is not particularly new knowledge, should be a lot of online content,

What you want to combine their practical error, as long as the data source is accurate enough, enough data, you don't have to optimization can also be right in theory,

In the learning stage, you can need not optimized, with the most common way to see,
With a little optimization, plus RANSANC look at effect how, for example,
After plus BA or other means

P. s. cheer

CodePudding user response:

refer to 6th floor Seasons_y response:

Quote: refer to 4th floor alittlenewbiek response:
because you have the world coordinates of a single image can also be, as long as there are multiple sets of points,

Didn't know it, but in general the world coordinates need more frames to write using more depending on the geometry of the relationship between the expressions of the loss,

A picture of the optimization can only guarantee that the effect of a transformation matrix is good,
But if your world coordinate and pixel coordinate itself must enough, also can be applied to the other picture transformation matrix,

As long as you can find the relation between, BA can application

P. s. written a book, the author zhang SLAM
And I can see multiple view geometry in CV

Eldest brother, you said zhang SLAM, didn't see ah, see only a xi 'an electronic science and technology, he is not home page ah what SLAM books

I smiled, mixed up with everyone else

14, visual SLAM in author