CMU-Perceptual-Computing-Lab/openpose ? reverse-engineered prompt

Build me a real-time pose estimation application based on the OpenPose library. I want to be able to feed it a video file or my live webcam stream and see it detect all the people in the scene. For each person, it should draw a skeleton overlay tracking their body, hands, face, and even their feet, just like in the examples I've seen.

The main goal is to get a working demo running. It should be able to process a video and either display the output live or save it as a new video file. I'd also like the option to save the detected keypoint data for each frame into a format like JSON. This looks like a C++ project, but I see it has Python bindings. I'm more comfortable with Python, so please set up the environment and create a simple script that I can run from the command line to test it. Make sure to handle installing all the dependencies like OpenCV and downloading the necessary models.