RANSAC设置：关于参数，基元和迭代的几个问题

RANSAC Settings: A few questions on parameters, primitives, and iterations-

RANSAC设置:关于参数,基元和迭代的几个问题 -

1. While segmenting a 3D point cloud,how the minimum number of support points per primitive is decided in RANSAC?

   在划分3D点云时,如何在RANSAC中确定每个基元的最小支撑点数量?

2. Furthermore, out of 5 primitives : Plane, Sphere, Cone, Cylinder, Torus which primitives should be selected and how?

   此外,在5个基元中:平面,球体,锥体,圆柱体,圆环应该选择基元以及如何选择?

3. How maximum number of iterations in decided in RANSAC?

   如何在RANSAC中确定最大迭代次数?

Reference: Schnabel, Ruwen, Roland Wahl, and Reinhard Klein. "Efficient RANSAC for Point‐Cloud Shape Detection." Computer graphics forum. Vol. 26. No. 2. Blackwell Publishing Ltd, 2007.

参考文献:Schnabel,Ruwen,Roland Wahl和Reinhard Klein。 “用于点云形状检测的高效RANSAC。”计算机图形论坛。卷。 26. No. 2. Blackwell Publishing Ltd,2007。

1 个解决方案

#1

1. The minimum number of support points per primitive depends on the kind of primitive itself: it is the minimum number of points required to fit a primitive and recover its parameters. In some cases this minimum number of points also depends the actual method used to recover the parameters of an instantiated primitive from support points.

   每个基元的最小支撑点数取决于基元本身的类型:它是拟合基元并恢复其参数所需的最小点数。在某些情况下,此最小点数还取决于用于从支持点恢复实例化基元的参数的实际方法。

   For instance:

   • Plane: 3 points are sufficient,

     飞机:3分就够了,

   • Sphere: 4 points are sufficient,

     球体:4分就足够了

   • Cylinder: 3 points would work (find the plane of the points and fit a circle, the axis of the cylinder is the normal of the plane going through the center of the circle),

     气缸:3点可以工作(找到点的平面并拟合一个圆,圆柱的轴是穿过圆心的平面的法线),

   • Cone: 4 points would work (find the plane of the first three points and fit a circle like before, find the slope of the cone by using the 4th point).

     圆锥:4个点可以工作(找到前三个点的平面并像之前一样拟合圆,通过使用第4个点找到圆锥的斜率)。

2. This would depend on what is expected to be found in the input point cloud: if there is no cone or torus, it would make sense to not try fitting cones or tori. Starting with planes only and then extending to sphere and cylinders would already be a good start.

   这将取决于在输入点云中预期会发现什么:如果没有锥形或圆环形,则不尝试装配锥形或圆环形是有意义的。从飞机开始然后延伸到球体和圆柱体已经是一个良好的开端。

3. The number of iterations is based on the desired confidence and an estimate or guess of the ratio of inliers (See: http://en.wikipedia.org/wiki/RANSAC#Parameters for the standard formula).

   迭代次数基于期望的置信度和内部比率的估计或猜测(参见:http://en.wikipedia.org/wiki/RANSAC#Parameters for standard formula)。