We are going to implement a pointcloud-based solution which could estimate 6D poses of objects found in the scene, given their .ply models. The scene is a hard drive, the pose estimation should, for instance, return the 6D pose of the reader there, given a .ply model of the reader and a point cloud input.
PCL already has some functionalities for this task, one has to dig in and find out whether or not they work, or if they do, to what extend. The task is not a real-time one, of course, so the pose estimation can naturally take some time, but it should take too much either. Since the object is a shiny one (metallic), the acquired pointcloud may not be the best (if the view is top-down), I am open for discussions on how to position the camera to acquire the best pointcloud for this task. To my experience, a bit of tilt angle helps, but let me hear from you what you think.
About the recuiterMember since Mar 14, 2020 Chetan
from Pest, Hungary