In 3D vision technology, the feature extraction of the surface of the workpiece is the premise and key of 3D reconstruction. However, the natural features often appear less obvious so the feature extraction is very difficult. The laser grid is often projected onto the surface of the workpiece so that the workpiece surface can be identified with definable characteristics. Aiming at the characteristics of laser grid mark images, we propose a method of pixel weights and hypothetical model pre-detection based on the RANSAC algorithm for line feature extraction. Experimental results show that this method can not only overcome the disadvantages of large computation and parameter sensitivity of the RANSAC algorithm, but also has good accuracy and robustness.