Research On The Calculation And Accuracy Of SFS Recursive Algorithm Of 3D Reconstruction

It's still a very difficult problem to reconstruct 3D surface geometry from one gray-level image in the field of image technology. It is indicated in the aspects of low accuracy of reconstructed surface, the result's divergence of radiation equation and easily interfered illumination. All these disadvantages show that SFS method is incomplete. Due to its special feature, which can restore 3D shape information just from one gray-level image, in recent years, shape from shading technology as a key research direction of computer vision is concerned increasingly and has been widely used in industry, agriculture, country defense, medicine, space technology, etc, especially.On the base of comprehensive summarization and analysis of current SFS research work, a recursive algorithm of 3D reconstruction based on minimal triangular that was proposed by our team has been researched, especially in the process of the solution. Character of solution and the reason of difficult to get accurate solutions are also introduced and a better solution is proposed. The accuracy of solution is enhanced remarkably by using the method. The factors which affect accuracy are researched and a better recursive path is found in the process of reconstruction, and the accuracy of the reconstruction are enhanced by using the method of averaging the results. At last, solution and reconstruction are realized by computer directly, and the feasibility of this algorithm is discussed by analyzing the error of reconstruction, which makes the algorithm more consummate.By the initial experiment, on the base of the gray-level image from reverse calculation rendering algorithm, with the recursive algorithm of 3d reconstruction, a better result has been got. Through the contrast between the result and the real data of this 3d object, the error of element reconstruction is no more than fifteen parts of pixel in a radius of less than 0.9, and the error is mainly determined by the imaging resolution of this gray-level image and the distance between the reconstruction point and initial iteration point.