Detection and pose measurement of underground drill pipes based on GA-PointNet++

Abstract

Drilling for gas extraction, a common method in coal mine gas control, involves tedious loading and uploading of drill pipes. This study aims to design a method for detecting and measuring pose drill pipes using point cloud data. We present an experimental platform for acquiring drill pipe point cloud data under various lights. Additionally, we propose a GA-PointNet + + model, enhanced with an adversarial generation network. The pose of the drill pipe was calculated from the segmented pipe and pin point clouds. Results indicate that the intersection-over-union (IoU) values for pipe and pin, based on GA-PointNet + + , are 0.824 and 0.472, respectively. Evaluating the model's performance in recognizing the pin using the ROC curve yielded an AUC of 0.87. The combination of GA-Pointnet + + and RGB-D camera was used to pose drill pipes, achieving an average accuracy of 82.5% under different lighting conditions. Under lighting conditions of 25–35 lx with an added diffuser film and 10–15 lx, the accuracy reaches 90%, with average distance errors of 1.4 cm and 2.5 cm, and average angle errors of 3.5° and 3.7°, respectively. This has significant implications for the use of LED lights in underground environments. Therefore, the proposed drill pipe pose measurement method is of great significance for the intelligentization of coal mine drilling operations.

Graphical Abstract