Pose Estimation of Drilling Pipe Using a 3D-Sensor Network

Abstract

In this thesis, a pose estimation system for drill pipe-ends has been made. The systemutilizes 6 sensor nodes containing a Kinect v2 RGB-D camera in addition to a Jetson TX2computer. Pipe detection is done on 2D images from each camera using Tiny-YOLOv3.A pair of point clouds for the box-end and pin-end are made based on these detections, inaddition to the depth map which the Kinect provides. Further segmentation is done, andpose estimation using ICP is performed.When the pipes are placed on the ground and placed vertical at a minimum distance of5.5m away from the sensors, the system achieves a mean positional accuracy of≈2.6cm,≈2.0cmand≈6.5cm. Further, the mean rotational deviation is≈0.8◦,≈3.5◦aroundthe x- and y-axis.i