In order to solve the influence of noise spot on extraction accuracy in the online structured light three-dimensional measurement, the center line of laser stripe was extracted by density clustering gray centroid extraction algorithm. The method consists of two stages: The pre-extraction of center line and the final extraction of center line. The pre-extraction stage realizes the simultaneous extraction of the center line of laser and spot. In the final extraction stage, the connectivity-based density clustering algorithm is used to preserve the laser centerline and eliminate the noise spot. In the simulation experiment stage, the image with the size of 600pixel×600pixel and the laser center line was denoised. The root mean square error and the running time of each point between the extracted result and the real center line were used as the inspection criteria. The results show that the root mean square error of high brightness anisotropic spot, high brightness small area spot, and high brightness point noise image was respectively reduced by 12.59pixel, 15.12pixel, and 83.36pixel, and the time complexity was respectively increased by 0.383s, 0.412s, and 0.416s. Compared with the traditional gray centroid method, this method has higher extraction accuracy, approximate time complexity, and better robustness to noise spot.