Underground rock bolt detection

Rock bolts are commonly used to reinforce ground in underground mining and civil tunnel environments. Rock bolts serve two main functions: i) they suspend large, loose blocks of ground, and ii) provide a protective pressure arch to restrain the deformation in an excavated void. Automated and precise tracking of rock bolt positions can assist with the operational success of ground support by reducing the amount of labour required in current manual practices. A point cloud is a large list of x, y and z coordinates in three-dimensional space, that when visualised, show a precise 3D representation of a physical environment. Point cloud data can be collected using accurate LiDAR scanners. These scanners can be mounted on a stationary tripod or on a moving vehicle. A key way point cloud data is used is in the bolt detection project being completed in conjunction with UNSW Sydney, which aims to provide an automated location of underground rock bolts from LiDAR scans. The immediate challenge with identifying rock bolts is that they constitute a very small portion of the data set amounting to approximately 0.01 per cent of some of the collected civil tunnel data. The process of detecting rock bolts from point clouds therefore requires a quick elimination of large background data while still preserving the bolt data. A further challenge is that the geometry of underground environments can often be very complex and non-uniform. To overcome these challenges, a two-step coarse to fine deep learning approach is taken. The first step is a coarse background elimination step aimed at removing as much of the background data as possible. The second step is a finer classification method, which uses deep learning to further segment the bolts from the remaining background. Additionally, a computer vision module is being implemented, which is expected to increase the detection rate by adding another dimension to the training data set.