A Proposal of Collation Method Using Nonlinear SVM for Restoration of Stone Wall in Kumamoto Castle

The Kumamoto Castle in Japan was disastrously damaged by the Kumamoto earthquakes in 2016. In this study, we focus on collapsed stone collation of the Kumamoto Castle. Stone wall in Kumamoto Castle is a important cultural facilities, it is required to find an accurate original position for each stone. In the previous research, contour information is used to collate the location of stones. This previous study showed that contour information was useful for stone wall collation. However, there are some problems, it was impossible to obtain complete contour information and stones with few features are difficult to identify correctly stone. Then, we were focused on the position of the fallen stones to improve the ability to find the correct original location in previous study. By using non-linear SVM that trained by fall position information, it is expected that we can narrow down collapse area of the collapsed stones compared to the previous study. The effectiveness is illustrated by numerical simulations. Moreover, we propose an stone wall collation algorithm by multimedia using fall position information in addition to contour information.