車載レンジセンサを用いた長大トンネルの3次元モデリングとその評価

Precise and accurate 3D model of a road structure is the basic information which can be utilized for various purposes, such as safety measures, 3D navigation, driving simulation, and reference data for autonomous driving. Generally, the 3D model of a tunnel where GPS signal is not available has been constructed based on positioning using gyro sensor, however, error accumulation becomes a considerable problem in the case of long tunnels. We propose a method to obtain geometrically optimal whole structure of a tunnel:(1) Acquire a set of partial structures by static scanning using a laser range sensor, which is mounted on the roof of a survey vehicle, and align them by 3D matching using edge feature. (2) Fix the absolute position of the data for both ends of the tunnel by GPS and align the rest data again. By applying this method, we succeeded to create the 3D digital model of Kanaya Tunnel in New Tomei Expressway, whose length is 4.6 km. For the comparison and evaluation of modeled tunnel, we present a system to evaluate the quality of the modeled tunnel comparing with the 2D CAD blueprint. This system can be described as following steps:(1)Extract the boundaries both from CAD blueprint and modeled tunnel to a 2D plane. (2)Transform the coordinate of modeled tunnel to the same coordinate as CAD blueprint. (3)Calculate the closest distance between the modeled tunnel boundaries and CAD boundaries. (4)Calculate the middle line of two data boundaries, and calculate the curvatures of both middle line, finally compare the curvature graph with the closest distance graph. We succeed creating a dense model of actual tunnel with a length of 4.6 km by geometric process. For the error evaluation, we did the curvature calculation and closest distance calculation. The closest distance of our modeled tunnel with 2D CAD data is about 50 centimeters on the west side, about 20 centimeters on the east side and about 4 meters around the big curve in the middle of tunnel. From this result ,we have three assumptions about the error factors, which can be described as follows, the uncertainty in central part, the rotation error around the big curve during the local alignment process, and the GPS positioning errors accumulated in the simultaneous alignment. From curvature analysis, we can see that the largest error is at the central part, and the error rate is about 0.1 percent, also the GPS points should not be fully fixed in global alignment. About our future work, we plan to increase our modeling accuracy and reduce errors at three parts. For data acquisition, we want to increase more density of our raw data, for local alignment, we want to use more other features besides edges for local alignment to reduces accumulated errors, lastly, for global alignment, we want to allow some freedom for GPS points to move slightly