[为了提高python脚本的效率,由于Numpy加快了处理过程,我试图基于点云上的许多“ for循环”操作来转换脚本。
概括地说,我有一个3d模型,表示为np.array中包含的一组3d点(x,y,z坐标)(维度:(188706,3))
[[168.998 167.681 0.] <-- point 1
[169.72 167.695 0.] <-- point 2
[170.44 167.629 0.] <-- point 3
...
[148.986 163.271 25.8] <-- point 188704
[148.594 163.634 25.8] <-- point 188705
[148.547 163.667 25.8]] <-- point 188706
[[[row x-1] [row x]的每对代表3D片段]
[[168.998 167.681 0.] [169.72 167.695 0.] <-- SEGMENT 1 (points 1 & 2)
[169.72 167.695 0.] [170.44 167.629 0.] <-- SEGMENT 2 (points 2 & 3)
[170.44 167.629 0.] [171.149 167.483 0.] <-- SEGMENT 3 (points 3 & 4)
...
[149.328 162.853 25.8] [148.986 163.271 25.8] <-- SEGMENT 94351 (points 188703 & 188704)
[148.986 163.271 25.8] [148.594 163.634 25.8] <-- SEGMENT 94352 (points 18874 & 188705)
[148.594 163.634 25.8] [148.547 163.667 25.8]] <-- SEGMENT 94353 (points 188705 & 188706)
我的目标是测量每个有序点//成对的行之间的欧几里得距离(=每个段的长度),以便我可以检测到需要添加更多点以表示更多3d模型表面的位置。换句话说,如果线段长度超过阈值(= 0.5mm),我将不得不用更多点离散化此特定线段,并将这些其他点添加到我的点云中。
由于此代码,我找到了一种方法来递归地测量每行之间的欧几里得距离:
EuclidianDistance = np.linalg.norm(PtCloud[:-1] - PtCloud[1:],axis=1)
哪个给出此结果:
[0.72213572 0.72301867 0.72387637 ... 0.54008148 0.5342593 0.05742822]
而且我还找到了如何根据段的顶点(末端)对段进行降序处理:
def AddEquidistantPoints(p1, p2, parts):
return np.stack((np.linspace(p1[0], p2[0], parts + 1), np.linspace(p1[1], p2[1], parts + 1)), axis=-1)
if EuclidianDistance > 0.5mm:
dist = AddEquidistantPoints(currentRow, previousRow, 10) #10 --> nb subdivisions
但是我的第一个问题是那些欧氏距离仅需在z坐标相等的点上计算即可。当z坐标不相等时,我是否要拆分数组?带有:
PtCloud = np.split(PtCloud, np.where(np.diff(PtCloud[:,2])!=0)[0]+1)
这给了我一个数组列表,所以我想我很遗憾会使用for循环...
而且我的第二个问题与递归检查和离散化步骤有关,因为我不知道如何在这种特殊情况下实现它。我想知道是否有没有任何for循环的方法。
因此,如果有人能帮助我解决这一难题,我将感到非常高兴,因为我目前被“困住了”。对我来说,这开始变得非常具有挑战性。
谢谢,埃尔维
与您分享,我刚刚找到一种解决我的问题的方法。这样做可能不是最有效的方法,但它可以起作用。
import numpy as np
print("====================")
print("Initial Points Cloud")
print("====================")
POINTCLOUD = np.array([[168.998, 167.681, 0.],
[169.72, 167.695, 0.],
[170.44, 167.629, 0.],
[171.149, 167.483, 0.],
[150.149, 167.483, 4.2],
[160.149, 167.483, 4.2],
[159.149, 166.483, 4.2],
[152.149, 157.483, 7.],
[149.328, 162.853, 25.8],
[148.986, 163.271, 25.8],
[148.594, 163.634, 25.8],
[180.547, 170.667, 25.8],
[200.547, 190.667, 25.8]])
print(POINTCLOUD)
print("============================================")
print("Reshaped Point Cloud in the form of segments")
print("============================================")
a = np.column_stack((POINTCLOUD[:-1],POINTCLOUD[1:]))
print(a)
b = a.reshape((a.shape[0],2, 3))
#print(b)
print("")
print("*******************************")
print("Non filtered euclidean distance")
print("*******************************")
EuclidianDistance = np.transpose(np.linalg.norm(b[:,0] - b[:,1],axis=1))
print(EuclidianDistance)
print("")
print("****************")
print("Mask computation")
print("****************")
mask = np.transpose([a[:,2] == a[:,5]])
mask2 = [a[:,:] == a[:,:]]*np.transpose([a[:,2] == a[:,5]])
print("")
print(mask2)
print("")
print("***********************************")
print("1rst Filter applyed on points cloud")
print("***********************************")
# b = np.where(mask2,a,0)
# b = np.squeeze(b, axis=0)
b = np.squeeze(np.where(mask2,a,0), axis=0)
print(b)
print("")
b2 = b[np.squeeze(mask2,axis=0),...].reshape((np.sum(mask),b.shape[1]))
print(b2)
print("")
#print(b2.reshape(b2.shape[0],2, 3))
b = b2.reshape(b2.shape[0],2, 3)
print("")
print("***************************************")
print("FIRST EUCLIDEAN DISTANCE FILTERING STEP")
print("***************************************")
EuclidianDistance = np.linalg.norm(b[:,0] - b[:,1],axis=1)
print(EuclidianDistance)
print("")
print("***************************")
print("# THRESHOLD MASK GENERATION")
print("***************************")
threshold = 7
mask_threshold = np.transpose(EuclidianDistance>threshold)
print(mask_threshold)
print("")
print("**********************************")
print("# FINAL FILTERED ECLIDEAN DISTANCE")
print("**********************************")
EuclidianDistance = EuclidianDistance[np.squeeze(mask_threshold),...]
print(EuclidianDistance)
print("")
print("**********************")
print("SEGMENTS TO DISCRETIZE")
print("**********************")
SegmentToDiscretize = b[np.squeeze(mask_threshold),...]
print(SegmentToDiscretize)
print("")
print("******************************")
print("EQUIDISTANT POINTS COMPUTATION")
print("******************************")
nbsubdiv2 = np.transpose(np.ceil(np.array(np.divide(EuclidianDistance,0.7))).astype(int)).reshape((SegmentToDiscretize.shape[0],1))
print(nbsubdiv2)
print(nbsubdiv2.shape)
print(nbsubdiv2[1,0])
print(SegmentToDiscretize.shape)
print(SegmentToDiscretize[:,0])
nbsubdiv = [10,30,10]
addedpoint = np.linspace(SegmentToDiscretize[:,0],SegmentToDiscretize[:,1],nbsubdiv[0], dtype = np.float)
addedpoint = addedpoint.reshape((addedpoint.shape[0]*addedpoint.shape[1],3))
print(np.sort(addedpoint,axis=0))
print("")
print("***********************************")
print("UPDATED POINT CLOUD WITH NEW POINTS")
print("***********************************")
# duplicates are removed with the command np.unique
POINTCLOUD = np.unique(np.append(POINTCLOUD,addedpoint, axis=0),axis=0)
print(POINTCLOUD)
print("")
print("************************")
print("FINAL SORTED POINT CLOUD")
print("************************")
sortedPOINTCLOUD = POINTCLOUD[np.argsort(POINTCLOUD[:, 2])]
print(sortedPOINTCLOUD)
print("***************************")
[如果需要,请随时添加自己的提案以进行改进。非常欢迎!