3d-reconstruction of destructive process models using remote sensing by a group of unmanned aerial vehicles

Abstract

The paper presents a novel method of volumetric reconstruction of transient destructive processes using remote sensing by a group of unmanned aerial vehicles. The study is based on the most common class of such processes like forest fires, where a fire front is a determinant, and its propagation reflects the dynamics of the process. The effects of wind, smoke and fire, turbulence and vibration, interference, distortion, and obstacles lead to uncertainty of observations, to overcome which fuzzy sets, soft sets and gray numbers were combined. A spatial model based on a recursive eight-fold subdivision of space as well as on a hierarchical structure of virtual cells is proposed, which allowed to resolve the contradictions between the accuracy and rate of reconstruction. The set of possible states of virtual cells is determined and the algorithm of their classification based on the use of a five-channel image recognition system containing infrared, two main, and two additional optical channels is proposed. An algorithm for calculating a 3D observation vector, presented by an array of confidence vectors, is proposed, which can be used to determine the gray fuzzy state of virtual cells allowing a combination of observations from different observers and refining them sequentially. The terrain where the process evolves is represented by a soft gray fuzzy set of virtual cells, which belong to a specific state at the consideration time, allowing identification of convincing, uncertain, suspicious, and negative components. The first one defines a stable core of the fire front while the second one represents its variation caused by uncertainty. The proposed method allows the reconstruction of transient spatially distributed processes of other classes, smoothing the effects of distortions and noise and ensuring acceptable performance.