RANSAC 随机采样一致性算法
RANSAC是一种随机参数估计算法。RANSAC从样本中随机抽选出一个样本子集,使用最小方差估计算法对这个子集计算模型参数,然后计算所有样本与该模型的偏差,在使用一个预先设定好的阈值与偏差比较,当偏差小于阈值时,该样本点属于模型内样本点(inliers),简称局内点或内点。否则为模型外样本点(outliers)。
然后迭代,取inliers的个数最多的作为结果返回;
PCL中Sample_Consensus支持的几种几何模型有:
- SACMODEL_PLANE:平面模型
- SACMODEL_LINE:直线模型
- SACMODEL_CIRCLE2D:二维圆的圆周模型
- SACMODEL_SPHERE:三维球体模型
- SACMODEL_CYLINDER:圆柱体模型
- SACMODEL_CONE:圆锥模型,尚未实现
- SACMODEL_TORUS: 圆环面模型,尚未实现
- SACMODEL_PARALLEL_LINE:有条件的直线模型,直线模型与给定轴线平行
- SACMODEL_PERPENDICULAR_PLANE:有条件限制的平面模型,平面模型与给定轴线垂直
- SACMODEL_NORMAL_PLANE: 有条件限制的平面模型,第一个局内点的发现必须与估计的平面模型的法线平行
- SACMODEL_PARALLEL_PLANE: 有条件限制的平明模型 在规定的最大角度偏差限制下,平面模型与给定的轴线平行
- SACMODEL_NORMAL_PARLLEL_PLANE:有条件限制的平面模型 三维平面模型与用户设定的轴线平行
下边的例子演示了
无参数时原始点云平面内点与立方体外点
参数为-f 提取得到的平面上的内点
参数为-s 提取得到圆球内点与立方体外点
参数为 -sf 提取圆球上的点为内点
无参数
参数为-f
参数为-s
参数为-sf
# -*- coding: utf-8 -*-
# How to use Random Sample Consensus model
# http://pointclouds.org/documentation/tutorials/random_sample_consensus.php#random-sample-consensus
import numpy as np
import random
import pcl.pcl_visualization
import math
import sys
argvs = sys.argv # for random model
# argvs = '-f' # plane model
# argvs = '-sf' # sphere model
argc = len(argvs) # 参数个数
print(argc)
print('argvs: ',argvs)
# 初始化点云
cloud = pcl.PointCloud()
final = pcl.PointCloud()
points = np.zeros((5000, 3), dtype=np.float32)
RAND_MAX = 1024
for i in range(0, 5000):
if argc > 1:
if argvs[1] == "-s" or argvs[1] == "-sf":
points[i][0] = 1024 * random.random() / (RAND_MAX + 1.0)
points[i][1] = 1024 * random.random() / (RAND_MAX + 1.0)
if i % 5 == 0:
points[i][2] = 1024 * random.random() / (RAND_MAX + 1.0)
elif i % 2 == 0:
points[i][2] = math.sqrt(math.fabs(1 - (points[i][0] * points[i][0]) - (points[i][1] * points[i][1])))
else:
points[i][2] = -1 * math.sqrt(
math.fabs(1 - (points[i][0] * points[i][0]) - (points[i][1] * points[i][1])))
else:
points[i][0] = 1024 * random.random() / RAND_MAX
points[i][1] = 1024 * random.random() / RAND_MAX
if i % 2 == 0:
points[i][2] = 1024 * random.random() / RAND_MAX
else:
points[i][2] = -1 * (points[i][0] + points[i][1])
else:
points[i][0] = 1024 * random.random() / RAND_MAX
points[i][1] = 1024 * random.random() / RAND_MAX
if i % 2 == 0:
points[i][2] = 1024 * random.random() / RAND_MAX
else:
points[i][2] = -1 * (points[i][0] + points[i][1])
cloud.from_array(points)
model_s = pcl.SampleConsensusModelSphere(cloud)
model_p = pcl.SampleConsensusModelPlane(cloud)
if argc > 1:
if argvs[1] == "-f":
ransac = pcl.RandomSampleConsensus(model_p)
ransac.set_DistanceThreshold(.01)
ransac.computeModel()
inliers = ransac.get_Inliers()
elif argvs[1] == "-sf":
ransac = pcl.RandomSampleConsensus(model_s)
ransac.set_DistanceThreshold(.01)
ransac.computeModel()
inliers = ransac.get_Inliers()
else:
inliers = []
else:
inliers = []
print(str(len(inliers)))
if len(inliers) != 0:
finalpoints = np.zeros((len(inliers), 3), dtype=np.float32)
for i in range(0, len(inliers)):
finalpoints[i][0] = cloud[inliers[i]][0]
finalpoints[i][1] = cloud[inliers[i]][1]
finalpoints[i][2] = cloud[inliers[i]][2]
final.from_array(finalpoints)
isWindows = True
if isWindows == True:
if argc > 1:
if argvs[1] == "-f" or argvs[1] == "-sf":
viewer = pcl.pcl_visualization.PCLVisualizering('3D Viewer')
viewer.SetBackgroundColor(0, 0, 0)
viewer.AddPointCloud(final, b'inliers cloud')
viewer.SetPointCloudRenderingProperties(pcl.pcl_visualization.PCLVISUALIZER_POINT_SIZE, 3, b'inliers cloud')
viewer.InitCameraParameters()
else:
viewer = pcl.pcl_visualization.PCLVisualizering('3D Viewer')
viewer.SetBackgroundColor(0, 0, 0)
viewer.AddPointCloud(cloud, b'sample cloud')
viewer.SetPointCloudRenderingProperties(pcl.pcl_visualization.PCLVISUALIZER_POINT_SIZE, 3, b'sample cloud')
viewer.InitCameraParameters()
else:
viewer = pcl.pcl_visualization.PCLVisualizering('3D Viewer')
viewer.SetBackgroundColor(0, 0, 0)
viewer.AddPointCloud(cloud, b'sample cloud')
viewer.SetPointCloudRenderingProperties(pcl.pcl_visualization.PCLVISUALIZER_POINT_SIZE, 3, b'sample cloud')
viewer.InitCameraParameters()
while viewer.WasStopped() != True:
viewer.SpinOnce(100)
else:
pass
版权声明:本文为qq_40985985原创文章,遵循 CC 4.0 BY-SA 版权协议,转载请附上原文出处链接和本声明。