The cloth simulation filter is important for the separation of features and is widely used in the field of point cloud classification. However,the ground point clouds obtained by this method often contain off-ground point clouds,which results in inaccurate outcome.The paper proposes an automatic optimization method for cloth simulation filter that considers the neighborhood features of the object,which can quickly and accurately solve this problem. Firstly,the data is filtered by cloth simulation filter. the off-ground point cloud of the results is segmented. A series of subsets are obtained；Secondly,fit the point cloud of each subset to obtain the main plane. Normalize the elevation of ground point clouds；Finally,filter ground point cloud by the third potential difference. Select three sets of representative data to test. Results showed that compared with the cloth simulation filter,the overall accuracy of the proposed was improved,especially in class II errors,and the accuracy was increased by an average of 12.3%,which can effectively solve the problem of the cloth simulation filter.