How to estimate pose of imported 3d model in pointcloud?
We want to estimate the pose of a single object using the ZED2 camera and ROS Melodic. My plan is to import a mesh-file (.stl for example) and search in the generated point cloud (made by the ZED2 camera) for the outline of the object. As we don't have sufficient and precise training data, our guess is, that the use of a CNN (Convolutional Neural Network) would not be suitable.
We were also thinking about writing our own algoriths, but it seems rather hard to check an RGBD-image or pointcloud for all different orientations of the imported object.
Are there any libraries that you can recommend for the pose estimation of single objects, that don't rely on neural networks with lot of training data?
I believe what you describe is called ICP: Iterative Closest Point.
There are many implementations of that available.
But please realise: this is not ROS-specific at all (so no use/need to limit yourself to that in your search queries).