Abstract: In the process of the multi-layer welding, the second layer is affected greatly by the shape and welding quality of the first layer, which subjects to the artificial observation and real-time adjustment of welding parameters. In order to meet the requirements of automatic surfacing, a precise and efficient 3D surface recognition technology is needed urgently. To solve this problem, a Delaunay triangulation model based on Kriging interpolation between three-dimensional data of sampling points is proposed. According to the data collected by magnetron rotary arc surfacing sensor, the data of the sampling points are smoothed and weighted by Kriging interpolation algorithm to obtain the corresponding interior points. Then the Delaunay triangulation of these data points with increasing division is performed to triangle mesh patches in the projected three-dimensional space to complete surfacing surface reconstruction. The results show that the method has a better surface reconstruction effect and higher real-time performance.