Line laser scanning measurement is a major area of interest within the field of 3D laser scanning measurement. Traditionally, sub-pixel extraction of laser stripes is a dominant point for line laser scanning measurement. In particular, the noise separation of laser stripe images and the accuracy of feature extraction of the laser stripe are the main challenges for sub-pixel extraction of laser stripes in complex circumstances. To this end, this study utilizes a robust and accurate method with two steps to extract sub-pixel features of laser stripes for 3D laser scanning measurement. Laser stripe segmentation based on a deep semantic segmentation network is initially implemented for noise elimination of images. Then, the sub-pixel extraction of the gray peak points of laser stripes is accomplished by Shepard sub-pixel interpolation and gray surface fitting, which can adequately utilize the gray distribution of laser stripes and obtain high-precision and anti-interference results. The robustness, effectiveness, and accuracy are verified by comparative experiments with classical methods. The results indicate that the proposed method can obtain much more complete, denser, and smoother results than traditional methods, especially in challenging measurement conditions, such as a large curved surface, a highly reflective surface, or intense ambient light. The accuracy of the proposed method can meet the requirements of high-precision measurement.