meiqua

这个矩阵是图像绕中心旋转init_angle角的矩阵，减去（opencv getRotationMatrix2D逆时针转）icp的旋转角就是最终角度了

You can refer to [this comment](https://github.com/meiqua/shape_based_matching/issues/12#issuecomment-474341789).

haven't tried C#. What's the time cost in C++ under the same condition?

c++ seems good. For c#, maybe c++ can't be used directly? There should be a standard way for c# to use c++ class.

Sorry, I'm not familiar with C#. But I think using same dll in C++ and C# the time should be similar.

信息太少，可以debug看看在哪崩溃

是的，fusion branch是fusion的时候直接copy出来的

``` c++ Detector(int num_features, std::vector T, float weak_thresh = 30.0f, float strong_thresh = 60.0f); ``` 可以调低阈值试试，第三个是匹配时的阈值，第4个是训练时的。

是的。一般来说，训练时想提到物体自己的边缘，所以设的高一点排除阴影之类的干扰。这里需要根据情况调小

刚好昨天ICP改进搞好了，可以参考[这个issue](https://github.com/meiqua/shape_based_matching/issues/100)，把后两个阈值调低点