3D radiation map reconstruction strategy using Gaussian process regression with attenuation effects and radiation characteristics

Recently, 3D radiation field reconstruction using data collected by robot has been examined to reduce radiation exposure to workers inside and outside nuclear facilities. Many studies have proposed radiation field reconstruction methods that consider radiation diffusion characteristics; however, they often face challenges in accounting for the effects of the surrounding environment. We proposed a novel kernel function of Gaussian process regression that considers both radiation characteristics and attenuation effects. Our kernel function incorporates diffusion properties and attenuation effects caused by obstacles. We extract geometry information from point cloud data and applied it to model the attenuation effects when building radiation map. To validate the novel kernel function, radiation data are collected from simulation and real-world datasets, then compared with radiation field estimation methods that do not consider attenuation effects. The experimental results show that in the attenuation regions, our proposed method achieved root mean square error reductions of 78.24 %, 43.65 %, and 69.26 % for the barrel environment, reactor environment, and real-world dataset, respectively, compared to the previous method. The proposed strategy demonstrated higher accuracy in predicting radiation maps by better reflecting the physical characteristics of real world. This capability enables rapid responses during incidents and helps minimize potential damage.