Pose2Art: smartCamera Pose to TouchDesigner etc
Pose2Art creates an AI Camera that can run on small 'edge' device that extract Pose and sends tracked points via OSC to TouchDesigner (and Unity/UnReal) for ART!
The project is documented in longer form in the Hackaday.io project Pose2Art: SmartCam to TouchDesigner, Unity via OSC. It is an evolving project, with the initial setups (esp pushing into Raspberry Pi) being a bit complex, so see that project for the how-to as well as discussion of details of the code, etc. As of June 2023, this repo is ahead of Hackaday project,
Album for Pose2Art demo videos
See related repo pyUdpTest for udp sender/receiver tests. These can be used to test both the udp connection, and as receiver of OSC messages. (TL:DR turn off windows defender)
There are a couple tools to implement the pose capture and sending OSC, along with a number of TouchDesigner applications that utilize the messages.
- pose_PC_MediaPipe.py: python tool for PC, sending 33 landmarks
- pose_rPi_TFLite.cpp: raspberryPi4 c++ program, sending 17 landmarks (in subfolder rPi_prototypes)
- posePCGui.py: Python TK gui over pose_detector; select webcam/file, ndi name, osc IP/port, loop video. Displays in TK window as well as sending osc and orig image via NDI
- pose_detector package (subfolder)
- pose_detector.py: PoseDetector base class
- pose_detector_mediapipe: PoseDetectorMediapipe version
- alphapose.py : stub that might support alphapose
Sample TouchDesigner apps:
- poseOSC_dots.toe is a simple TouchDesigner app that reads the osc landmark messages and displays them as dots.
- pose_OSC_Sender.py is a python app for testing OSC Messages. Initially it sends 17 or 33 landmarks plus the Frame messages. A sendBundle() is provided to test if receivers can handle OSC Bundles.
- handDrawing.toe: revision of example from HQPro to use OSC instead of kinect (part of paid course) -- Immersive Design & Creative Technology Mini-Degree/Pillars/Kinect 2 Fundamentals/05 - Drawing with Kinect Skeleton Data
- handDrawing_wNDI.toe: TD app maps over ndi from posePCGui
- landmarksAsSpheres.toe puts a sphere at each landmark (incl face?)
- landmarksAsGeom_wNDI.toe similar to sphere, also rcvs video via NDI
- handsAsEmitters.toe particle emitters at hand tips (vid save to file)
- OSC_TubularTrim.toe make flat/tube skeleton between landmarks
- osc_fluidHand.toe fluid emitter on hands using two Fluid_simulation.tox
The testScripts folder holds scaffolding test tools
- oscServer.py: pythonosc server that prints received messages
- send_capture.py: NDI example with minor changes
- testMediaPipe.py: tests PoseDetectorMediapipe with webcam
- testMediaPipeFile.py: tests PoseDetectorMediapipe with video file
- udpServerPrint.py: udp socket server prints data received
A number of explorations are NOT in git at this time. Mostly these are various TouchDesigner apps that explore different effects off the skeleton. I've also been looking into alternative Pose2Art cameras that can do multiple person and/or Hand Tracking and Gestures recognition.
Critique
Initially there were 2 hardware SmartCamera for Pose_OSC: a PC's webcam and raspberryPi4. The PC can do a respectable frame rate with decent graphics board. My PC webcam gets about 28fps, which is decent. The rPi4 about 8-9fps, which is not good enough for interactive work. Maybe the Jetson Nano or Coral dev boards will be better. Meanwhile, we at least have a path for getting ML data into TouchDesigner via OSC. This method could be extended for multiple person tracking (on fast hardware), object detection or other ML processing. The OSC messages will need to change for those new applications, so when you fork this, document them ;-)
Credit Where Credit Due
The initial code is derived from other projects on GitHub/web.
- pose_PC_MediaPipe.py and TouchDesigner poseOSC_dots.toe come from cronin4392's TouchDesigner-OpenCV-OSC project (MIT License)
- pose_rPi_TFLite files come from Q-Engineering's rPi image projects (BSD 3-Clause License)
- Fluid Simulation relies on Bruno Imbrizi's youTube tutorial and Kurt Kaminski's uploading of that as a TOX on Community.Derivative. Kurt has a revised version on GitHub