Problems encountered in the construction of tunnels are most of the time due to fractured rock. In this semester project, we propose an approach to reconstruct the fracture network in cataclastic rock samples based on X-ray computer tomograms.
The first step of the algorithm consists in surface detection using steerable filters. Surface detection is generally a computationally intensive task since it involves the computation of the optimal orientation of the surface detector and its corresponding response. Therefore, steerable filters are ideally suited for our application since they can be very efficiently rotated by taking a linear combination of basis filters.
The detection procedure is then followed by the suppression of non-maxima values and a hysteresis thresholding step. However, the results at this point show that the fractures have been fragmented and that a certain amount of noise is remaining.
In order to have a consistent representation of the fracture network, methods reducing the amount of detections caused by the noise in the original volume, have been developed. Finally, we used an approach based on Dijkstra algorithm to reconnect the fractures.