基于多视角热像图序列的物体表面温度场重建OA北大核心CSTPCD

The surface temperature field model effectively reflects the temperature distribution on the surface of the object.Since the image information collected from one single viewpoint cannot cover the entire object,it is difficult to reconstruct the temperature field of the entire object surface.Additionally,in the temperature field model,the presence of non-target objects can affect temperature model analysis.A method for reconstructing the object's surface temperature field based on a multi-view thermal image sequence was proposed.Firstly,a semantic segmentation algorithm was employed to extract the contours of the target object from visible light images,and then the fusion of the single view temperature point cloud of the target object was achieved by combining depth data with temperature information from the thermal image.Subsequently,the method of multi-view image data and multi-view temperature point cloud stitching was adopted,and the initial stitching of the temperature point cloud was carried out using the poses of cameras from various viewpoints,the multi-view LM-ICP algorithm was applied to optimize the registration of the global temperature point cloud.Experimental results show that the method effectively reconstructs the object surface temperature field model,and has small size error(2.61 mm)and temperature errors(0.56℃).