三维点云学习(2)下-Octree
三维点云学习(2)下-Octree
Octree(八叉树)的基本结构
Octree(八叉树)的构建
Better one:
如果有一个球体完全包围一个正方体,那只需要对相应的octant进行kNN search就行
图示:
如果:
图示绿色线 < radius 可认为被包裹
复杂度
完整代码
Radius search normal:
Search takes 45438.469msRadius search fast:
Search takes 20330.457ms
import random
import math
import numpy as np
import timefrom lidar_KnnRnnClass_2 import KNNResultSet, RadiusNNResultSetclass Octant:def __init__(self, children, center, extent, point_indices, is_leaf):self.children = childrenself.center = centerself.extent = extentself.point_indices = point_indicesself.is_leaf = is_leafdef __str__(self):output = ''output += 'center: [%.2f, %.2f, %.2f], ' % (self.center[0], self.center[1], self.center[2])output += 'extent: %.2f, ' % self.extentoutput += 'is_leaf: %d, ' % self.is_leafoutput += 'children: ' + str([x is not None for x in self.children]) + ", "output += 'point_indices: ' + str(self.point_indices)return outputdef traverse_octree(root: Octant, depth, max_depth):depth[0] += 1if max_depth[0] < depth[0]:max_depth[0] = depth[0]if root is None:passelif root.is_leaf:print(root)else:for child in root.children:traverse_octree(child, depth, max_depth)depth[0] -= 1def octree_recursive_build(root, db, center, extent, point_indices, leaf_size, min_extent):if len(point_indices) == 0:return Noneif root is None:root = Octant([None for i in range(8)], center, extent, point_indices, is_leaf=True)# determine whether to split this octantif len(point_indices) <= leaf_size or extent <= min_extent:root.is_leaf = Trueelse:root.is_leaf = Falsechildren_point_indices = [[] for i in range(8)]for point_idx in point_indices:point_db = db[point_idx]morton_code = 0if point_db[0] > center[0]:morton_code = morton_code | 1if point_db[1] > center[1]:morton_code = morton_code | 2if point_db[2] > center[2]:morton_code = morton_code | 4children_point_indices[morton_code].append(point_idx)# create childrenfactor = [-0.5, 0.5]for i in range(8):child_center_x = center[0] + factor[(i & 1) > 0] * extentchild_center_y = center[1] + factor[(i & 2) > 0] * extentchild_center_z = center[2] + factor[(i & 4) > 0] * extentchild_extent = 0.5 * extentchild_center = np.asarray([child_center_x, child_center_y, child_center_z])root.children[i] = octree_recursive_build(root.children[i],db,child_center,child_extent,children_point_indices[i],leaf_size,min_extent)return rootdef inside(query: np.ndarray, radius: float, octant:Octant):"""Determines if the query ball is inside the octant:param query::param radius::param octant::return:"""query_offset = query - octant.centerquery_offset_abs = np.fabs(query_offset)possible_space = query_offset_abs + radiusreturn np.all(possible_space < octant.extent)def overlaps(query: np.ndarray, radius: float, octant:Octant):"""Determines if the query ball overlaps with the octant:param query::param radius::param octant::return:"""query_offset = query - octant.centerquery_offset_abs = np.fabs(query_offset)# completely outside, since query is outside the relevant areamax_dist = radius + octant.extentif np.any(query_offset_abs > max_dist):return False# if pass the above check, consider the case that the ball is contacting the face of the octantif np.sum((query_offset_abs < octant.extent).astype(np.int)) >= 2:return True# conside the case that the ball is contacting the edge or corner of the octant# since the case of the ball center (query) inside octant has been considered,# we only consider the ball center (query) outside octantx_diff = max(query_offset_abs[0] - octant.extent, 0)y_diff = max(query_offset_abs[1] - octant.extent, 0)z_diff = max(query_offset_abs[2] - octant.extent, 0)return x_diff * x_diff + y_diff * y_diff + z_diff * z_diff < radius * radiusdef contains(query: np.ndarray, radius: float, octant:Octant):"""Determine if the query ball contains the octant:param query::param radius::param octant::return:"""query_offset = query - octant.centerquery_offset_abs = np.fabs(query_offset)query_offset_to_farthest_corner = query_offset_abs + octant.extentreturn np.linalg.norm(query_offset_to_farthest_corner) < radiusdef octree_radius_search_fast(root: Octant, db: np.ndarray, result_set: RadiusNNResultSet, query: np.ndarray):if root is None:return Falseif contains(query, result_set.worstDist(), root):# compare the contents of the octantleaf_points = db[root.point_indices, :]diff = np.linalg.norm(np.expand_dims(query, 0) - leaf_points, axis=1)for i in range(diff.shape[0]):result_set.add_point(diff[i], root.point_indices[i])# don't need to check any childreturn Falseif root.is_leaf and len(root.point_indices) > 0:# compare the contents of a leafleaf_points = db[root.point_indices, :]diff = np.linalg.norm(np.expand_dims(query, 0) - leaf_points, axis=1)for i in range(diff.shape[0]):result_set.add_point(diff[i], root.point_indices[i])# check whether we can stop search nowreturn inside(query, result_set.worstDist(), root)# no need to go to most relevant child first, because anyway we will go through all childrenfor c, child in enumerate(root.children):if child is None:continueif False == overlaps(query, result_set.worstDist(), child):continueif octree_radius_search_fast(child, db, result_set, query):return Truereturn inside(query, result_set.worstDist(), root)def octree_radius_search(root: Octant, db: np.ndarray, result_set: RadiusNNResultSet, query: np.ndarray):if root is None:return Falseif root.is_leaf and len(root.point_indices) > 0:# compare the contents of a leafleaf_points = db[root.point_indices, :]diff = np.linalg.norm(np.expand_dims(query, 0) - leaf_points, axis=1)for i in range(diff.shape[0]):result_set.add_point(diff[i], root.point_indices[i])# check whether we can stop search nowreturn inside(query, result_set.worstDist(), root)# go to the relevant child firstmorton_code = 0if query[0] > root.center[0]:morton_code = morton_code | 1if query[1] > root.center[1]:morton_code = morton_code | 2if query[2] > root.center[2]:morton_code = morton_code | 4if octree_radius_search(root.children[morton_code], db, result_set, query):return True# check other childrenfor c, child in enumerate(root.children):if c == morton_code or child is None:continueif False == overlaps(query, result_set.worstDist(), child):continueif octree_radius_search(child, db, result_set, query):return True# final check of if we can stop searchreturn inside(query, result_set.worstDist(), root)def octree_knn_search(root: Octant, db: np.ndarray, result_set: KNNResultSet, query: np.ndarray):if root is None:return Falseif root.is_leaf and len(root.point_indices) > 0: #如果是末尾节点(leaf)那就直接把数据扔进KNNResult# compare the contents of a leafleaf_points = db[root.point_indices, :]diff = np.linalg.norm(np.expand_dims(query, 0) - leaf_points, axis=1)for i in range(diff.shape[0]):result_set.add_point(diff[i], root.point_indices[i])# check whether we can stop search nowreturn inside(query, result_set.worstDist(), root)# go to the relevant child firstmorton_code = 0if query[0] > root.center[0]:morton_code = morton_code | 1if query[1] > root.center[1]:morton_code = morton_code | 2if query[2] > root.center[2]:morton_code = morton_code | 4if octree_knn_search(root.children[morton_code], db, result_set, query):return True# check other childrenfor c, child in enumerate(root.children):if c == morton_code or child is None:continueif False == overlaps(query, result_set.worstDist(), child):continueif octree_knn_search(child, db, result_set, query):return True# final check of if we can stop searchreturn inside(query, result_set.worstDist(), root)def octree_construction(db_np, leaf_size, min_extent):N, dim = db_np.shape[0], db_np.shape[1]db_np_min = np.amin(db_np, axis=0)db_np_max = np.amax(db_np, axis=0)db_extent = np.max(db_np_max - db_np_min) * 0.5db_center = db_np_min + db_extentroot = Noneroot = octree_recursive_build(root, db_np, db_center, db_extent, list(range(N)),leaf_size, min_extent)return rootdef main():# configurationdb_size = 64000dim = 3leaf_size = 4min_extent = 0.0001k = 8db_np = np.random.rand(db_size, dim)root = octree_construction(db_np, leaf_size, min_extent)# depth = [0]# max_depth = [0]# traverse_octree(root, depth, max_depth)# print("tree max depth: %d" % max_depth[0])# query = np.asarray([0, 0, 0])# result_set = KNNResultSet(capacity=k)# octree_knn_search(root, db_np, result_set, query)# print(result_set)## diff = np.linalg.norm(np.expand_dims(query, 0) - db_np, axis=1)# nn_idx = np.argsort(diff)# nn_dist = diff[nn_idx]# print(nn_idx[0:k])# print(nn_dist[0:k])begin_t = time.time()print("Radius search normal:")for i in range(100):query = np.random.rand(3)result_set = RadiusNNResultSet(radius=0.5)octree_radius_search(root, db_np, result_set, query)# print(result_set)print("Search takes %.3fms\n" % ((time.time() - begin_t) * 1000))begin_t = time.time()print("Radius search fast:")for i in range(100):query = np.random.rand(3)result_set = RadiusNNResultSet(radius = 0.5)octree_radius_search_fast(root, db_np, result_set, query)# print(result_set)print("Search takes %.3fms\n" % ((time.time() - begin_t)*1000))if __name__ == '__main__':main()
三维点云学习(2)下-Octree相关推荐
- 三维点云学习(1)下-点云体素降采样
三维点云学习(1)下 点云体素降采样(Voxel Filter Downsampling) 代码参考网址秦乐乐CSDN博客 理论参考知乎博主:WALL-E 1.方法 Centroid 均值采样 Ran ...
- 三维点云学习(2)五种算法比较
三维点云学习(2)五种算法比较 代码参考来自 黎老师github 本次测试包含五种算法比较: octree print("octree --------------")#时间统计c ...
- 三维点云学习(3)7- 实现GMM
三维点云学习(3)7- 实现GMM github大神参考代码 高斯混合模型的通俗理解 GMM课程个人总结笔记 最终效果图 原图 进行高斯聚类后的图 代码编写流程 1.输入数据集x1 x2 -xn,和K ...
- 三维点云学习(3)2- K-Means
三维点云学习(3)2- K-Means K-Means的具体构建流程 将输入的N个数据点,分为N个类 1.随机选取K个中心点 2.E-Step(expectation):N个点.K个中心,求N个点到K ...
- 三维点云学习(3)1-聚类数学理论
三维点云学习(3)1-数学理论 Linear Algebra 谱定理 对于一个对称矩阵,我们可以用他的特征值特征向量的线性组合表示 瑞利熵 Probability – Joint Probabilit ...
- 三维点云学习(1)上-PCA主成分分析 法向量估计
三维点云学习(1)上 环境安装 1.系统环境 win10 或者 ubuntu 2. Anaconda3+python3.6 使用Anaconda创建的conda虚拟环境进行python的编写 环境安装 ...
- 三维点云学习(5)5-实现Deeplearning-PointNet-2-classfication
三维点云学习(5)5-实现Deeplearning-PointNet-2-classfication Github PointNet源码 数据集下载:为40种物体的三维点云数据集 提取码:es14 运 ...
- 三维点云学习(5)4-实现Deeplearning-PointNet-1-数据集的批量读取
三维点云学习(5)4-实现Deeplearning-PointNet-1-数据集的批量读取 Github PointNet源码 数据集下载:为40种物体的三维点云数据集 提取码:es14 因为本人初次 ...
- 三维点云学习(5)3-Deep learning for Point Cloud-PointNet++
三维点云学习(5)3-Deep learning for Point Cloud-PointNet++ 强烈推荐 PointNet的深刻理解 PointNet++示意图 可分为 Encoder. Se ...
最新文章
- HIVE-ORC表一些知识点
- java 入门 第三季1
- java中日期与字符串之间的转换
- [WCF安全系列]绑定、安全模式与客户端凭证类型:NetNamedPipeBinding、NetTcpBinding与NetMsmqBinding...
- oracle10g配置失败6,Oracle10g DataGuard中ORA-16026错误解决-入门基础-Oracle频道-中国IT实验室...
- MACOSX下查看某个端口被哪个程序占用及杀进程方法
- docker 安装hadoop
- swoole task 异步任务 注释请忽略 自己加的不一定对 别误导大家。。。。。。
- linux UID的长度,linux下文件数、目录数、文件名长度的各种限制
- android电视自动关机,Android定时关机问题解决
- 联想g510拆键盘的简单方法_联想g510笔记本电脑键盘拆卸视频
- QOS中PQ,CQ.RR,WFQ,CBWFQ,LLQ区分
- SEO基础知识简介(一)
- JAVA:JDBC数据库编程
- python代码示例大全 下载-python基础代码大全
- CT与DR双能X射线物质识别算法实现与应用(工业选煤、稀土分拣、毒爆检测、垃圾分类等)
- 【QT Creator学习记录】(一)上位机与下位机串口通信
- 为了搞清楚类加载,竟然手撸JVM!
- .nc地形数据的python转换实现tiff
- opencvsharp历程_opencvsharp 例程
热门文章
- NFS客户端挂载目录后无写入权限的解决方案
- linux 下mysql忘记密码或者安装好linux后不知道mysql初始密码解决方案
- IIS配置Url重写实现http自动跳转https的重定向方法(100%解决)
- 活动子项父项的复杂CSS选择器[重复]
- 如何在JavaScript中将浮点数转换为整数?
- hdfs mv命令_Hadoop2.x HDFS shell命令
- 教你win7显卡驱动如何安装,win7电脑显卡驱动的安装方法
- Linux,vi编辑器使用手册
- mysql centos7安装_Linux——CentOS7之mysql5.7安装与配置
- Android:复杂listview条目