I was looking into free to use options for tracking. I found MediaPipe Hand Landmarks to be excellent for identifying hands, however I wanted something more specific for feet tracking, so that a 3D model can be placed on detected feet and follow them. After some research and forums, I saw that MediaPipe Pose Landmarks is better for feet.
I couldn’t find any sample repositories though, so I was hoping for some help regarding which MediaPipe solution is better between the 2, if tracking can be implemented within three.js and/or if there any sample GitHub repositories I could use to understand how to set that up.
MediaPipe Hand Landmarks is excellent, but it is really optimized only for hands. For feet tracking, MediaPipe Pose is the better choice since it provides full body landmarks, including ankles, heels, and foot index points, which makes it much easier to attach and drive a 3D model on the feet.
Another good option is using TensorFlow or TensorFlow.js with a full body pose model such as MoveNet or BlazePose. These models also expose reliable foot landmarks and work well in browser based setups.
Both MediaPipe Pose and TensorFlow pose models can be integrated with three.js by mapping the detected landmark positions to your scene coordinates and updating a mesh or rig in real time. This approach is commonly used for placing and following 3D objects on tracked body parts.
At the moment there are not many clean, feet focused sample repositories, but looking at general MediaPipe Pose or TensorFlow pose plus three.js demos is usually enough to understand the setup and adapt it for foot tracking.