Vision 3D measurement is an important non-destructive testing method in pipeline machining and performance monitoring. In order to apply the measurement system to the restricted space in the pipeline and ensure measurement accuracy, the existing 3D measurement methods mainly focus on the study of the measurement principle and the sensor structure, there is a lack of research on pipeline image processing and feature extraction algorithms, and the metal material of pipeline leads to the problem of reflection in the image, which seriously affects the accuracy of feature extraction and pipeline measurement. In order to study the feature extraction of reflective surfaces in vision 3D measurement of the pipeline, a method of reflection removal based on dynamic region segmentation and a method of laser center extraction are proposed. A dynamic annular mask is designed according to the morphological characteristics of the projected laser of the measurement system to eliminate the reflection interference, and an automatic selection method of Gauss kernel and extremum points is designed to overcome the interference of uneven feature width in the laser center extraction. The accuracy of the method is verified by experiments, and the characteristic coordinates are provided for the vision 3D measurement of the pipeline. The diameter and coaxiality of the pipeline are measured, and the measurement accuracy is higher than 0.1 mm.