Considering the problems of complex parameter setting, cumbersome threshold setting, and low registration accuracy in the traditional point cloud feature extraction method, a geometric feature extraction method based on the surface feature degree of point cloud is proposed. First, the normal vector and curvature of the point cloud are determined according to the weighted covariance matrix and combined with both the angle and curvature feature of the normal vector to form the surface feature degree. The median value is used as the threshold to automatically screen the feature points. Then, based on the improved dung beetle optimization algorithm (ST-DBO), the six parameters of point cloud registration are automatically optimized. The ST-DBO algorithm enhances the global optimization ability, which can prevent falling into a local optimal solution and determine the optimal parameters of the registration. Finally, through comparison experiments on the two public data sets and the actual collected point cloud data, the experimental results show that the registration accuracy of the proposed algorithm is improved by up to 33.12% compared to that of the typical traditional point cloud feature extraction method. Moreover, the feature extraction efficiency is significantly improved, such that it can adapt to different point cloud data. Additionally, the registration accuracy is improved by 29.13% compared to those with other registration algorithms. Under different parameter settings and noise interferences, the proposed algorithm exhibits strong robustness, good feature extraction efficiency, and improved registration accuracy.