cpd配准存在的问题

奇怪的缩放

%降采样
movingDownsampled = pcdownsample(moving,‘gridAverage’,0.0025);
fixedDownsampled = pcdownsample(fixed,‘gridAverage’,0.0025);
如下图所示：

原因分析：
cpd配准算法在计算某个点的概率的时候是基于两点之间的距离值的，所以对于变形较大的位置，可能出现就近匹配的情况，导致错误的匹配，程序就出现部分与整体匹配或者缩放错误的情况。

较好的结果

%降采样
movingDownsampled = pcdownsample(moving,‘gridAverage’,0.005);
fixedDownsampled = pcdownsample(fixed,‘gridAverage’,0.0025);
如下图所示：

分析：在fixed的点比较多，而moving的点比较少的时候，大部分点的匹配效果比较好，可是变形较大的地方的匹配效果还是不好。

混乱的匹配

%降采样
movingDownsampled = pcdownsample(moving,‘gridAverage’,0.0025);
fixedDownsampled = pcdownsample(fixed,‘gridAverage’,0.005);
如下图所示：

分析：在moving的点比较多，而fixed的点比较少的时候，整体效果较差。

最后的尝试

try1:
%降采样
movingDownsampled = pcdownsample(moving,‘gridAverage’,0.0035);
fixedDownsampled = pcdownsample(fixed,‘gridAverage’,0.0025);
如下图所示：

try 2:
%降采样
movingDownsampled = pcdownsample(moving,‘gridAverage’,0.004);
fixedDownsampled = pcdownsample(fixed,‘gridAverage’,0.0025);
如下图所示：

分析：看上去效果较好。
结论：
匹配的效果与降采样的策略有很大的关系，不具有泛化性。