casychow / iterative-closest-point Goto Github PK

Hi @casychow , Thanks for your fantastic work. I've tested several ICPs, and yours seems the most stable. However, I have some questions about several lines in your code. Could you please give some explanations to help me understand?

1. L39-L40, it seems there is an initial transformation for the source points, if I comment this init transformation, the registration is much worse, but I think it should not be such case (I also tested with my data, it still has similar performance which is weird, why this initial transformation can be applied to a custom data?). I know that the better the initial alignment, the better the results. Why and how do you use such an initial transformation?

2. L61. Here you compute the correspondence, could you give some detailed explanation on this? Why can we get the correspondence from this ?

Looking forward to your reply, thanks!