K-means 方法实现3D点云的聚类算法

import numpy as np
import open3d as o3d
import copy
import random
from matplotlib import pyplot as plt
from sklearn import datasets
from mpl_toolkits.mplot3d import Axes3D
#matplotlib notebook##draw labels on the point cloud
def draw_labels_on_model(pcl,labels):cmap = plt.get_cmap("tab20")pcl_temp = copy.deepcopy(pcl)max_label = labels.max()colors = cmap(labels / (max_label if max_label > 0 else 1))pcl_temp.colors = o3d.utility.Vector3dVector(colors[:,:3])o3d.visualization.draw_geometries([pcl_temp])d = 4
mesh = o3d.geometry.TriangleMesh.create_tetrahedron().translate((-d, 0, 0))
mesh += o3d.geometry.TriangleMesh.create_octahedron().translate((0, 0, 0))
mesh += o3d.geometry.TriangleMesh.create_icosahedron().translate((d, 0, 0))
mesh += o3d.geometry.TriangleMesh.create_torus().translate((-d, -d, 0))
mesh += o3d.geometry.TriangleMesh.create_moebius(twists=1).translate((0, -d, 0))
mesh += o3d.geometry.TriangleMesh.create_moebius(twists=2).translate((d, -d, 0))##apply k means on this point cloud
point_cloud = mesh.sample_points_uniformly(int(1e3))
##transfer point cloud into array
xyz = np.asarray(point_cloud.points)##define several necessary methods here
#normalize the dataset
def normalize(X,axis=-1,p=2):#normalize the array and then transfer into a vectorlp_norm = np.atleast_1d(np.linalg.norm(X,p,axis))lp_norm[lp_norm == 0] = 1#expand a dimension along axis for lp_norm#this is to make sure X and lp_norm have the same dimensionsreturn X / np.expand_dims(lp_norm,axis)def euclidean_distance(one_sample,X):#transfer one_sample into 1D vectorone_sample = one_sample.reshape(1,-1)#transfer X into 1D vectorX = X.reshape(X.shape[0],-1)#this is used to make sure one_sample's dimension is same as Xdistances = np.power(np.tile(one_sample,(X.shape[0],1))-X,2).sum(axis=1)return distancesclass Kmeans():#constructordef __init__(self,k=2,max_iterations=1500,tolerance=0.00001):self.k = kself.max_iterations = max_iterationsself.tolerance = tolerance#randomly select k centroidsdef init_random_centroids(self,X):#save the shape of Xn_samples, n_features = np.shape(X)#make a zero matrix to store valuescentroids = np.zeros((self.k,n_features))#bcs there is k centroids, so we loop k tiemsfor i in range(self.k):#selecting values under the range radomlycentroid = X[np.random.choice(range(n_samples))]centroids[i] = centroidreturn centroids#find the closest centroid of a sampledef closest_centroid(self,sample,centroids):distances = euclidean_distance(sample,centroids)#np.argmin return the indices of the minimum of distancesclosest_i = np.argmin(distances)return closest_i#determine the clusersdef create_clusters(self,centroids,X):n_samples = np.shape(X)[0]#This is to construct the nested list for storing clustersclusters = [[] for _ in range(self.k)]for sample_i, sample in enumerate(X):centroid_i = self.closest_centroid(sample,centroids)clusters[centroid_i].append(sample_i)return clusters#update the centroids based on mean algorithmdef update_centroids(self,clusters,X):n_features = np.shape(X)[1]centroids = np.zeros((self.k,n_features))for i, cluster in enumerate(clusters):centroid = np.mean(X[cluster],axis=0)centroids[i] = centroidreturn centroids#obtain the labels#same cluster, same y_pred valuedef get_cluster_labels(self,clusters,X):y_pred = np.zeros(np.shape(X)[0])for cluster_i, cluster in enumerate(clusters):for sample_i in cluster:y_pred[sample_i] = cluster_ireturn y_pred#predict the labelsdef predict(self,X):#selecting the centroids randomlycentroids = self.init_random_centroids(X)for _ in range(self.max_iterations):#clustering all the data pointclusters = self.create_clusters(centroids,X)former_centroids = centroids#calculate new cluster centercentroids = self.update_centroids(clusters,X)#judge the current difference if it meets convergence  diff = centroids - former_centroidsif diff.any() < self.tolerance:breakreturn self.get_cluster_labels(clusters,X) if __name__ == "__main__":clf = Kmeans(k=6)labels = clf.predict(xyz)draw_labels_on_model(point_cloud,labels)

K-means 方法实现3D点云的聚类算法相关推荐

毫米波雷达点云 DBSCAN聚类算法
毫米雷达点云 DBSCAN聚类算法聚类的目的聚类算法分类原型聚类层次聚类密度聚类 DBSCAN聚类算法原理相关定义算法流程以及伪代码 DBSCAN算法优缺点 DBSCAN参数选择聚类衡 ...
PCL 点云DBSCAN聚类算法
文章目录一.简介二.实现代码三.实现效果参考资料一.简介关于DBSCAN聚类算法的原理可以看之前的文章:点云DBSCAN聚类算法(C++),这里仅仅是对该算法进行了整理,让它可以支持PCL ...
3D点云补全算法汇总及最新进展
点击上方"3D视觉工坊",选择"星标" 干货第一时间送达 Part 1 前言在探讨3D 点云补全专题前,先介绍三个概念: 概念一:partial obser ...
3D-BoNet：比3D点云实例分割算法快10倍！代码已开源
点击我爱计算机视觉标星,更快获取CVML新技术本文转载自新智元(AI_era) 新智元报道来源:投稿编辑:元子 [新智元导读]本文提出了一种基于边界框回归的高效点云实例分割算法,通过最小 ...
【点云论文速读】点云分层聚类算法
点云PCL免费知识星球,点云论文速读. 标题:PAIRWISE LINKAGE FOR POINT CLOUD SEGMENTATION 作者:Lu, Xiaohu and Yao, Jian and ...
点云DBSCAN聚类算法（C++）
文章目录一.DBSCAN算法二.算法步骤三.代码实现效果四.小结一.DBSCAN算法 DBSCAN算法,全称为"Density-Based Spatial Clustering o ...
多任务上实现SOTA，UBC、谷歌联合Hinton等提出3D点云的无监督胶囊网络
点击上方"3D视觉工坊",选择"星标" 干货第一时间送达作者丨杜伟.小舟来源丨机器之心这是一种为 3D 点云提出的无监督胶囊架构,并且在 3D 点云重构. ...
UBC、谷歌联合Hinton等提出3D点云的无监督胶囊网络，多任务上实现SOTA
本文转载自机器之心. 机器之心报道作者:杜伟.小舟这是一种为 3D 点云提出的无监督胶囊架构,并且在 3D 点云重构.配准和无监督分类方面优于 SOTA 方法. 理解对象是计算机视觉的核心问题之一 ...
UBC、谷歌联合Hinton等提出3D点云的无监督胶囊网络，多任务上实现SOTA | AI日报...
多任务上实现SOTA,UBC.谷歌联合Hinton等提出3D点云的无监督胶囊网络理解对象是计算机视觉的核心问题之一.传统方法而言,理解对象任务可以依赖于大型带注释的数据集,而无监督方法已经消除了对标 ...

K-means 方法实现3D点云的聚类算法

K-means 方法实现3D点云的聚类算法相关推荐

最新文章

热门文章