baojunshan/pose3d2bvh

body25 + hand pose3d to bvh

pose3d to bvh

这个repo主要是把 openpose中body25格式，或者body25加上手部姿势的keypoints，转换成bvh。

目前已有的转换大多是h36的，在此更新body25和手势。本项目可以帮助那些使用openpose或者body25作为骨骼结构的同学，做bvh的粗略生成。

注：3d keypoints转换bvh只能做到大概的姿势相似，可以通过相对位置计算子节点相对于父节点的欧拉角，但是，对于人体骨骼而言，其自旋仅凭keypoints是无法或者很难精确算出的，需要其他先验知识，在这里就不展开了。

This repo mainly converts the body25 format in openpose, or body25 plus the key points of the hand posture, into bvh.

Most of the repos in github are about h36 skeleton, thus I update the body25 one. This project can help guys who use openpose or body25 skeleton as a tool to do rough generation of bvh.

Note: 3d keypoints converting into bvh can only achieve approximate pose similarity, but the relative position calculation can only calculate the Euler angle of the parent node. For human skeleton, relatively speaking, keypoints are unrealistic to convert to bvh exactly, such as the spin angle. It requires other prior knowledge to assist, and the detials will not be expanded here.

pose3d2bvh's Issues

Standard pose skeleton

Hello!

First of all, thank you for your great work :)

I am struggling to understand the creation of the standard skeleton here

pose3d2bvh/bvh/body25hand21.py

Line 28 in 2022457

self.con2dir = {

More specifically, I am struggling to understand what is the thought process for the hands.

For example, why is (7, 30): [3, 0, 1] picked here?

pose3d2bvh/bvh/body25hand21.py

Line 79 in 2022457

(4, 51): [-3, 0, 1], # w->index_finger

Hope you can help, thank you in advance!

动捕数据

请问你这3DPOSE数据是自己生成的么，我用的单目摄像头做的3DPOSE数据很不稳，显示出来很晃。

骨骼自旋的问题

师兄您好，我看了github上一些pose转bvh的项目中都会用到您写的类似
x_dir = None
y_dir = pose[index['heel_r']] - pose[joint_idx]
z_dir = pose[joint_idx] - pose[index['knee_r']]
order = 'zxy'
这样的代码，这个和您提到的

对于人体骨骼而言，其自旋仅凭keypoints是无法或者很难精确算出的，需要其他先验知识，在这里就不展开了。

这个是不是有关系，这个x_dir，y_dir，z_dir的设定以及后续的代码我理解的是一个求解新的局部坐标系的过程，然后来据此算出方向余弦矩阵，请问几个*_dir的设计是什么原理呀，我有点理不太清，按照上面那个例子的意思是说ZoY平面始终是由这两段关节组成的这个平面嘛，这个是可以肯定的吗，不会有不一样的情况吗，希望您能给我简单讲一下，麻烦您了，谢谢！