Deep Learning Based Flame Surface 3D Reconstruction Algorithm Research

Flame is an important feature of the combustion process,and combustion has accompanied the birth of mankind,and still plays an important role in various aspects of social life,production,transportation,aerospace,etc.,either providing heat,energy,or power.The flame has the characteristics of high brightness,fast edge change,irregular surface area,etc.,which produces large errors using conventional 3D reconstruction algorithms.Due to the complexity of the combustion process,the coupling effect of multiple factors,and the limitation of experimental observation means,there have been many challenges and difficulties in its research.In order to solve the influence of background noise on the reconstruction results when reconstructing the flame surface of laminar flow flame,and at the same time improve the reconstruction accuracy of the flame surface,this paper proposes a deep learning-based multi-view 3D reconstruction network model(IM-MVSNet)for reconstructing the flame surface of laminar flow flame.The network obtains high-quality segmented images by image segmentation of the reference frames and neighboring frames of the input sampled images to remove the background noise during sampling,then depth estimation of the multiview images is performed to obtain the depth map,and then the segmented reference images are used to refine the depth map for 3D reconstruction,and finally a 3D point cloud of the laminar flame surface is constructed to obtain the reconstructed laminar flame surface.And a method to calculate the flame volume and area based on the 3D point cloud data of the flame surface is proposed.The method constructs the 3D point cloud data reconstructed from the multi-view flame data as a triangular grid model of the flame surface,and calculates the projected volume of a single triangular grid,and the area of each triangular grid,and then obtains the flame volume and area data.The main contributions of the paper are as follows.(1)A flame surface 3D reconstruction dataset for deep learning of the flame surface 3D reconstruction model is produced based on the multi-view stereo matching technique,which provides the necessary data support for conducting a deep learning based 3D reconstruction model of the flame surface.(2)A deep learning network for flame surface 3D reconstruction of laminar flow flames is proposed,which generates a clear shape of the flame surface depth map and can effectively reflect the flame morphology and help to carry out flame volume calculation.(3)A method is proposed to quickly calculate the flame volume as well as the area based on the 3D mesh data of the flame surface,and the application of the 3D reconstruction of the flame surface is initially explored.The calculation comparison results of different reconstruction models show that the 3D reconstruction network proposed in this paper can effectively reduce the background noise of the point cloud data obtained from the 3D reconstruction of flame surface,improve the accuracy of flame surface reconstruction,and obtain a clear shape of the flame surface point cloud data,showing the flame volume and area results from the flame image to the triangular grid of flame surface,and the calculated flame volume and area.The accuracy of the method is verified by calculating the volume and area of the object easily,which shows that the proposed method can calculate the flame volume and area easily and with high accuracy.It can provide a new technical tool for flame combustion quality assessment in the internal flow channel of aero-engine test,flame situation assessment at the fire site and turbulent flame research.