In recent years, dynamic projector-based augmented reality has attracted much attention. In this paper, we propose a technique to detect deformable surfaces with high degrees of freedom and project images in real time. Projection mapping techniques for currently deformable objects have marker recognition errors due to light effects and microscopic shading or noise issues. This causes a problem that the target and the projected image can not be precisely matched. Thus, the accuracy of marker detection generally determines the performance of the system. In this paper, we aim to improve the accuracy of marker detection and to apply it in real industry by enhancing the characteristics of markers and comparing them with previously learned features. We designed a marker with four distinct patterns. To recognize the deformable surface over time, we reconstruct the meshes with vertices of the patterns and map the texture to the reconstructed meshes to reinterpret the 3D surface recognition problem as a two dimensional mesh generation problem.