中文摘要：

光度立体技术是一种非接触式获取物体表面几何形状信息的重要方法,由表面法向量场进行表面形状重建是光度立体技术的关键环节.对现有方法的分析表明,三角网格算法只能实现局部重建且抗噪性能较差.为提高重建精度,引入类似于法切向法中的全局约束条件,提出了改进的三角网格法.利用朗伯体半球面模型,对法切向法、泊松法、三角网格法和改进的三角网格法的重建精度和计算时间进行比较.实验表明：在理想情况下,泊松法的重建时间较短,改进的三角网格法重建精度更高;在有噪声情况下,改进的三角网格法在重建精度和抗噪性能方面的表现都比较好.

英文摘要：

Photometric stereo is an important contactless method for obtaining object shape information. It is a key step in surface reconstruction based on surface normal field.By reviewing and analyzing existing methods, we find that performance of the triangle mesh method is unsatisfactory in local computation and vulnerable to noise. To improve reconstruction accuracy, an improved triangle mesh method with global constraints as used in the normal and tangent vector method is proposed. An object with Lambertian surface is used in the experiments to compare reconstruction accuracy and computing time of normal and tangent vector method, Poisson method, triangle mesh method and the improved triangle mesh method. The results show that Poisson method is computationally efficient, the improved triangle mesh method is better in reconstruction accuracy when the image does not contain noise, while the proposed triangle mesh method performs better both in reconstruction accuracy and noise resistance.