ryo-simon-mf / DDA-PoseNet

Posenet → OSC

*Used Posenet to OSC

This repository contains a modified version of PoseNet by Dan Oved that sends detected keypoints as OSC encoded datagram packets. This enables tracked keypoints to be used within any OSC compatible environment (Max/MSP, Python, PD, C++, Processing, you name it). This is effectively PoseNet + osc-js.

PoseNet is a wabcam-based real-time motion tracking system that runs in a browser using TensorFlow.

Pose Detection in the Browser: PoseNet Model

PoseNet can be used to estimate either a single pose or multiple poses, meaning there is a version of the algorithm that can detect only one person in an image/video and one version that can detect multiple persons in an image/video.

Refer to this blog post for a high-level description of PoseNet running on Tensorflow.js.

Installation

To get this running stand alone on your computer, clone this repo and follow these steps:

Install node (using homebrew):

brew install node

Install yarn:

brew install yarn

Install dependencies and prepare the build directory:

yarn

Watch files for changes, and launch a dev server:

yarn watch

In another terminal window run the bridge node application:

node bridge.js

In Firefox or Chrome, browse to http://localhost:1234

Now keypoints can be received as OSC messages on port 9876.

OSC messages will have an address pattern /pose/n/part, where n is the index of the pose (person) found in a frame, and all keypoints for that pose will follow as arguments in the order x, y where part is the keypoint body part (string) and x and y are the coordinates of the part (floats).

Keypoints

All keypoints correspond to a body part. The parts are:

Part
nose
leftEye
rightEye
leftEar
rightEar
leftShoulder
rightShoulder
leftElbow
rightElbow
leftWrist
rightWrist
leftHip
rightHip
leftKnee
rightKnee
leftAnkle
rightAnkle