Fringe Projection Profilometry is one of the important techniques in three-dimensional vision. However, traditional Fringe Projection Profilometry employing a lot of patterns has significant challenges for efficiently recovering the absolute phase. In this paper, a multi-purpose neural network combining with code-based patterns is proposed to recover the absolute phase firstly, which can greatly decrease the number of patterns with high accuracy. Different from the traditional approach of trigonometric calculation, the well-trained multi-purpose network can learn the principle of extracting absolute phase from few patterns. Experiments demonstrate that the proposed method can acquire high accuracy for complex texture objects, indicating potential application in precision engineering with high speed.

条纹投影轮廓仪是三维视觉中的重要技术之一。然而，采用许多图案的传统边缘投影轮廓仪对于有效地恢复绝对相位具有重大挑战。本文提出了一种结合基于代码的模式的多用途神经网络来首先恢复绝对相位，从而可以大大减少模式的数量，并且具有较高的精度。训练有素的多用途网络与传统的三角计算方法不同，可以学习从很少的模式中提取绝对相位的原理。实验表明，该方法对复杂的纹理物体具有较高的精度，在高速精密工程中具有潜在的应用前景。