What is the theory for SmallBlurryImage::IteratePosRelToTarget() method ?

[cggos]

I'm learning the PTAM code, but I cannot find the theory respond to the SmallBlurryImage::IteratePosRelToTarget() method linking https://github.com/Oxford-PTAM/PTAM-GPL/blob/master/Src/SmallBlurryImage.cc#L98-L205
I would appreciate it if you can tell me the theory or its thesis name.
I am looking forward to your reply, thank you.

[GeorgKlein]

This is described by section 6 of the ECCV 2008 paper "Improving the agility of keyframe-based SLAM". In short, we run ESM (similar to Lucas-Kanade) to align two images to each other over SE(2).

[cggos]

@GeorgKlein
I can not understand the code segment below in SmallBlurryImage::IteratePosRelToTarget() method.

1. SE2<> se2XForm = se2WfromC * se2CtoC * se2WfromC.inverse();
   what does se2WfromC and se2CtoC mean and why make se2XForm in that form ?

2. v4Jac[2] = -(ir.y - irCenter.y) * v2SumGrad[0] + (ir.x - irCenter.x) * v2SumGrad[1];
   what does v4Jac[2] mean and why compute it like this ?

3. dMeanOffset -= v4Update[3];
   why use v4Update[3] to make dMeanOffset and compute the dFinalScore ?

[GeorgKlein]

1. WfromC maps a image-center-based coordinate frame to an image-corner-based coordinate frame, as required by the warp function. 2. This looks like the partial differential of image intensity w.r.t. rotation about the image center, computed by multiplying the differential of image motion w.r.t. rotation by the differential of image intensity w.r.t. image motion. 3. The optimization runs over three parameters: (0&1): translation (2): rotation (3): intensity shift between images.

…

On Tue, Feb 6, 2018 at 12:55 AM, Gordon Scott ***@***.***> wrote: @GeorgKlein <https://github.com/georgklein> I can not understand the code segment below in SmallBlurryImage:: IteratePosRelToTarget() method. 1. SE2<> se2XForm = se2WfromC * se2CtoC * se2WfromC.inverse(); what does *se2WfromC* and *se2CtoC* mean and why make *se2XForm* in that form ? 2. v4Jac[2] = -(ir.y - irCenter.y) * v2SumGrad[0] + (ir.x - irCenter.x) * v2SumGrad[1]; what does *v4Jac[2]* mean and why compute it like this ? 3. dMeanOffset -= v4Update[3]; why use *v4Update[3]* to make *dMeanOffset* and compute the *dFinalScore* ? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#16 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AFftUXSIGsjvSBxyqEN8CxR8rXJPgwAHks5tSBOZgaJpZM4Rq9qi> .