The point cloud captured by millimeter wave radar, TOF camera, and other scanning equipment contains a large number of redundant information and noise points, which reduces the accuracy and processing efficiency of the point cloud registration step. In view of this phenomenon, a 3D point cloud simplification method based on KFCM is proposed. The algorithm maps the original scanned point cloud to feature space by the kernel function. At the same time, the clustering center and membership matrix were updated, and the objective function was established by using the sum of weighted error squares as the convergence mark of the algorithm. Finally, the clustering center was output as the simplified result. Experimental results show that the proposed method can reduce the resolution of source point cloud to achieve simplification effect while maintaining the basic characteristics of point cloud well, and the density of output point cloud can be controlled by adjusting parameters, which can meet the requirements of ICP and other registration algorithms in the adaptability experiment of registration task.