Improve motion convert calculations

Questions
,
1

Hello to all developers. I am creating an application for capturing movements and converting them into rotations. I want to improve the quality of 3d landmarks conversion.

Please advise what can be done, I use mmd models for animation. The conversion logic is here: Akira-MMD/src/app/ui/logic/MotionModel.ts at master · GOH23/Akira-MMD · GitHub

You can test it here: https://akirammd.vercel.app/

I take into account every suggestion you make.
I would be grateful if you help.
The main problems:

  • The moveBody method does not change the location along the x axis and produces conversion inaccuracies.
1 Like
3

Sorry for not providing the information properly(

4

Your project looks interesting! For improving motion conversion, you might want to refine how you handle offsets along the x-axis—maybe add a correction factor or interpolation for smoother transitions. Have you tried using Inverse Kinematics (IK) to improve accuracy? I’ll check out the demo and share more thoughts!

1 Like
5

I’m sorry I answered you late. The method corrects the position of the model’s molars based on the detected body landmarks. It calculates the center between the shoulders and hips to determine the midpoint of the body. Using the average height of the ankles and predefined scaling factors, it calculates the target position of the molar (for example, “センター” or “全の親”). The position is smoothly interpolated using the LERP coefficient to ensure natural movement. Could you rewrite one of the methods using inverse kinematics for understanding, please?

6

I fixed moveBody method

1 Like