Matlab中去除多余点的快速方法

Question

我在 Matlab 中合并了两个 pointCloud 对象，比如说 pc1 和 pc2。 pc1 是参考云，即 pc2 中与 pc1 中的点相等或非常接近的所有点都需要在组合云之前删除。

说明：

• 我知道函数 pcmerge，它几乎可以做我想做的事 - 但我肯定需要删除冗余点并且平均这些点不是一个选项

• 每个点云的大小约为 500,000，我必须比较其中的许多 (100) 个。这就是为什么速度很重要。

• 我希望能够围绕 pc1 的每个点定义一个半径，以给出“冗余”的标准。但是为了提高速度，一些简化是可以的（参见我的第二种解决方案）。

解决方法：

• 一个可行但非常缓慢的解决方案是在 pc2 中查找其最近邻居的每个点：

  function [ pc ] = pcaddcloud( pc1, pc2, res )
  
  limits = overlapRange(pc2, pc1);
  pc1idx = findPointsInROI(pc2, limits);
  pc2Overlap = select(pc2, pc1idx);
  idx = findPointsInROI(pc1, limits);
  pc1Overlap = select(pc1, idx);
  endi = pc2Overlap.Count;
  pc2Overlap = pc2Overlap.Location;
  for i=1:endi
      [idx, ~] = findNeighborsInRadius(pc1Overlap, pc2Overlap(i,:), res);
      % keep only indices of redundant points to delete them later
      if isempty(idx)
          pc1idx(i) = 0;
      end
  end
  pc1idx(pc1idx==0) = [];
  pc2 = pc2.Location;
  pc2(pc1idx,:) = [];
  pc = pointCloud([pc1.Location; pc2]);
  end
  
  % Compute the bounding box of overlapped region (from pcmerge)
  function rangeLimits = overlapRange(pcA, pcB)
  xlimA = pcA.XLimits;
  ylimA = pcA.YLimits;
  zlimA = pcA.ZLimits;
  
  xlimB = pcB.XLimits;
  ylimB = pcB.YLimits;
  zlimB = pcB.ZLimits;
  
  if (xlimA(1) > xlimB(2) || xlimA(2) < xlimB(1) || ...
          ylimA(1) > ylimB(2) || ylimA(2) < ylimB(1) || ...
          zlimA(1) > zlimB(2) || zlimA(2) < zlimB(1))
      % No overlap
      rangeLimits = [];
  else
      rangeLimits = [ min(xlimA(1),xlimB(1)), max(xlimA(2),xlimB(2)); ...
          min(ylimA(1),ylimB(1)), max(ylimA(2),ylimB(2)); ...
          min(zlimA(1),zlimB(1)), max(zlimA(2),zlimB(2))];
  end
  end
  

• 我有一个更快的解决方案（仍然很慢，但比解决方案 1 更快）处理 alpha 形状：我在 pc1 周围定义了一个外壳，并确定 pc2 的点是否在内部。缺点：仅“稍微向外”的点（即靠近 pc1 的点但不在 alpha 形状内）不会被检测为冗余。

  function [ pc ] = pcaddcloud( pc1, pc2 )
  
  limits = overlapRange(pc2, pc1);
  pc2 = pc2.Location;
  pc1 = pc1.Location;
  %seems to be faster than findPointsInROI:
  pc2Overlap = pc2(pc2(:,1)>=limits(1,1)&pc2(:,1)<=limits(1,2) ...
      &pc2(:,2)>=limits(2,1)&pc2(:,2)<=limits(2,2)...
      &pc2(:,3)>=limits(3,1)&pc2(:,3)<=limits(3,2),:);
  pc2idx = find(pc2(:,1)>=limits(1,1)&pc2(:,1)<=limits(1,2) ...
      &pc2(:,2)>=limits(2,1)&pc2(:,2)<=limits(2,2)...
      &pc2(:,3)>=limits(3,1)&pc2(:,3)<=limits(3,2));
  pc1Overlap = pc1(pc1(:,1)>=limits(1,1)&pc1(:,1)<=limits(1,2) ...
      &pc1(:,2)>=limits(2,1)&pc1(:,2)<=limits(2,2)...
      &pc1(:,3)>=limits(3,1)&pc1(:,3)<=limits(3,2),:);
  
  shape = alphaShape(double(pc1Overlap));
  in = inShape(shape, double(pc2Overlap));
  pc2idx(~in) = [];
  pc2(pc2idx,:) = [];
  pc = pointCloud([pc1; pc2]);
  
  end
  
  % Compute the bounding box of overlapped region (from pcmerge)
  function rangeLimits = overlapRange(pcA, pcB)
  xlimA = pcA.XLimits;
  ylimA = pcA.YLimits;
  zlimA = pcA.ZLimits;
  
  xlimB = pcB.XLimits;
  ylimB = pcB.YLimits;
  zlimB = pcB.ZLimits;
  
  if (xlimA(1) > xlimB(2) || xlimA(2) < xlimB(1) || ...
          ylimA(1) > ylimB(2) || ylimA(2) < ylimB(1) || ...
          zlimA(1) > zlimB(2) || zlimA(2) < zlimB(1))
      % No overlap
      rangeLimits = [];
  else
      rangeLimits = [ min(xlimA(1),xlimB(1)), max(xlimA(2),xlimB(2)); ...
          min(ylimA(1),ylimB(1)), max(ylimA(2),ylimB(2)); ...
          min(zlimA(1),zlimB(1)), max(zlimA(2),zlimB(2))];
  end
  end
  

我期待您的想法！如果需要，请随时询问更多信息 - 我是这个平台的新手。谢谢！

Answer 1

您可以使用ismembertol 和ByRows 选项来检测冗余点。但考虑到它使用立方邻域而不是球形邻域。 假设你有两个矩阵pc1,pc2，每个矩阵有3列和一个容差tol：

idx = ismembertol(pc2, pc1, tol,'ByRows', true, 'DataScale' , 1);
result = [pc1; pc2(~idx,:)];