Feactures related to the selection of KNN

[xiaoxin05]

hi，First of all，thank you for sharing. I have some problems about choosing feature？
i found that your used （x2-x1),(y2-y1）/（|x2-x1| + |y2-y1|）as the feature. why this can distinguish between different pose ? I think this aboved parameters cound't represent.

[RobertGCNiu]

You can check this part in the code. Actually, I used the dot product of two vector to distinguish different poses.

[xiaoxin05]

You can check this part in the code. Actually, I used the dot product of two vector to distinguish different poses.

hi，I am very happy to receive your reply.
Since I have seen your pose diagram, I thought before that how can the parameters (（x2-x1),(y2-y1）/（|x2-x1| + |y2-y1|） ) be distinguished, so I would like to ask you.
thanks.

I have been working on pose classification based on key points recently. Do you have any suggestions?

[RobertGCNiu]

I have tried Euclidean distance and vector angle between the training vector and test vector. I found the vector angle has a better performance. The recognition accuracy is larger than 90%. I have no idea to improve this project. Nevertheless, if you want to try, deep learning can be a potential approach.

[xiaoxin05]

Hello, sorry for the late reply. saw it very early actually, but really do not know why cannot reply, gradually delayed.
I really appreciate your prompt reply and many suggestions.