Inquiry about some codes in EpipolarSearch::compute function

[JicenHu]

Hi, Developer. Thanks for your great code. I have a question there in your epipolar_search.cpp:
In the void EpipolarSearch::compute(POI2D poi)* function:
"
view1_vector << (poi->x + poi->deformation.u), (poi->y + poi->deformation.v), 1;
Eigen::Vector3f view2_epipolar = fundamental_matrix * view1_vector;
...
int x_view2 = (int)((line_slope * (poi->y + poi->deformation.v + parallax.y - line_intercept) + poi->x + poi->deformation.u + parallax.x) / (line_slope * line_slope + 1));
..."
I am confused why in these two codes we should add poi->deformation.u and v in the calculation. If the deformation.u and v is close to the ground truth and the deformation itself is large, then your epipolar line would not correspond to your POI(x,y) rather than (x+u,y+v) in your reference point. Same thing applies for the x_view2 calculation where both deformation and parallax added seems redundant and might cause issue? I am considering whether the deformation.u and v part should be removed from these two mentioned codes.

If there is anything I misunderstand, Please help to revise me.

Thanks,
Jicen

[vincentjzy]

Hi Jicen,
Sorry for the delayed response due to intensive conferences and teaching in the past two weeks. The addtion of deformation (disparity in stereovision) is to facilitate the registration between the reference POI in primary camera view (moment 0) and the target POI in secondary camera view (moment i).

view1_vector locates the position of POI in primary camera view at moment i, thus the deformation with respect to the position of POI at moment 0 needs to be incorporated. Afterwards, the epipolar in the secondary camera view at moment i could be correctly estimated.

More details could be found in our paper
https://doi.org/10.1016/j.optlaseng.2021.106812

Please feel free to ask if you have any further quesion.

All the best,
Zhenyu