Accurate 6D Pose estimation using known 3D CAD model
HI
Hi. I have to estimate a very accurate 6D Pose of regular shape objects ( like blocks for construction ) when the RGBD Camara is placed by a distance of 1m from the object. The accuracy of the detected 6D Pose(position and orientation of the block) must be within 1cm (max 2-3cm). I find this Surface matching algorithm and the code surface matching But now sure if these can meet the accuracy requirement and also must work in real-time.
Also, would I like to know if is this code can solve the problem when the objects are one on top of the other? So when objects are identical and one on top of others but with a different orientation, and I have to get the orientation and the position of the object on the top. Would it be working in that case?
If not is there any other algorithm that can get used to the known 3D CAD model and the 2D image or Point clouds that is very accurate and in real-time and can estimate the 6D Pose when the objects are on top of the other?
Also, I find this D2CO: Direct Directional Chamfer Optimization. This software implements the D2CO (Direct Directional Chamfer Optimization) algorithm, a fast edge-based registration technique for accurate 3D object pose estimation. This one is using the 3D CAD model and 2D Images. Any help if this one can solve my problem?
Please, any help? Will really appreciate it
Thanks
any chance you can get 2d correspondances for your 3d cad model points ?
you could use solvePnP() which nicely works realtime
So one approach is to compare the 3DCAD model with 2D images and the other one compares 3D CAD model with the corresponding Point Clouds. I have to choose which one is more accurate. I need precision or around 1cm max 2-3cm by 1m, Im not sure how to get 2d correspondences for 3d cad model points. Would that lead to a very accurate 6D Pose estimation?
how to get the 2d correspondances for your 3d cad model points? Do i need to train the model ? Or just use orb or surf features?