结合深度学习的格雷码相位展开方法  

作　　者：庞琦 

机构地区：[1]东南大学自动化学院,江苏南京210096

出　　处：《工业控制计算机》2022年第7期69-70,72,共3页Industrial Control Computer

摘　　要：在基于格雷码进行相位展开过程中,如不能对格雷码进行准确的二值化则会影响最终的展开结果。然而由于全局照明等因素的影响,造成一部分像素点在经历暗条纹时的亮度高于经历亮条纹的亮度,因此对于基于强度的格雷码,正确的二值化仍旧是一个挑战。在该实验中,首先使用深度学习的方法对不同频率的格雷码进行二值化并得到绝对相位,然后使用传统方法得到绝对相位,通过分离直接光和间接光,将传统方法中间接光强于直接光的像素替换为基于深度学习的结果,该方法兼顾了传统方法和基于深度学习的方法,仿真实验证明该方法具有可行性和准确性。Failure to accurately binarise a Gray code during phase unwrapping can affect the final result.However,due to global illumination and other factors,there are some pixels that are brighter when they experience dark stripes than when they experience bright stripes,so it is still a challenge to binarise the intensity-based Gray codes.In this experiment,this paper firstly uses the deep learning method to binarise Gray codes with different frequencies and obtain the absolute phase,then use the traditional method to obtain the absolute phase.By separating direct and indirect light,the pixels where indirect light is stronger than direct light in the conventional method are replaced with the results based on deep learning,this method takes into account both the traditional method and the deep learning-based method,and simulation experiments prove the feasibility and accuracy of this method.

关 键 词：三维形貌测量 格雷码 全局照明 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]