In mountain tunnel excavation work, numerous accidents have been reported due to rock mass falls from a part of face surface. It is known that face surface suddenly displaces by a few millimetres or a few centimetres just before the collapse. Detecting such a small displacement in real-time enables urgent evacuation of construction workers for safety. However, it is very difficult for workers to find it visually. In this paper, we propose a new approach for real-time detection of sudden small displacement using high-speed 3D laser scanner. We used Velodyne HDL-32e scanner in this work. Displacement can be simply detected by comparing time-series scans. To compute the displacement accurately, a depth image is created, which is called base frame, by cylindrical expansion of sequential scans. At each pixel, it computes the averaged distance to the target of a set of points in corresponding pixels. Then for each consecutive frame, it calculates the differences of the distance to the base frame. Compared to the previous works using multiple laser displacement sensors or terrestrial laser scanner, our approach has several advantages. The scanning area is large enough so that it can capture the entire face surface, and the installation of physical targets to the surface is not required. In addition, the scanning frequency of HDL-32e is high up to 20 Hz, and thus is suitable for real-time detection. It is reported that HDL-32e contains time-varying nature, i.e., the range measurement results to the static objects vary in the long scanning periods. Specific calibration technique is proposed for this problem, however, they need large planar or cylindrical objects in a scene. Such objects do not exist in tunnel construction site, and carrying such objects is not easy in practice. Thus we skipped the calibration. We carried out several experiments. Firstly, for evaluating repeatability of the scanner, we scanned a flat wall in the scene multiple times and evaluated the variations of the points. As a result, variation is small and the average of the standard deviations of distances computed at each pixel was 4.3mm. The face surface mainly consists of rock and clay. To imitate the real face surface, we prepared the equipment which can be slide within 1mm accuracy by screw and mounted rocks and cray on it. We scanned this target before and after the slides from 20m distance and compared the difference. The averaged error of the detection accuracy of the slides was -1.96mm, and its standard deviation was 2.79mm when scanned from the front. Finally, we scanned the static target from the fixed position in the long periods (15 hours), and evaluated the difference of the measurement distances. Although the distances vary in the early stage of the scanning (within 1 hour), the static condition of the target was accurately captured by our approach, i.e., the differences of the consecutive frames were small enough. These results show some potentials of the proposed approach to be used in real tunnel construction site.
Laser scanning and photogrammetry , Automation and robotics for construction
