Developer Ethan Dell has released a video showcasing how to get started using TensorFlow Lite to run a pose estimation model on a Raspberry Pi 4 single-board computer, building on earlier work by Evan Juraes.
"[Pose estimation is] a technique that uses machine learning models to determine the pose of a person — basically how they're standing — and this is done through using machine learning techniques to determine where key body part points are," Dell explains.
"The goal of this project was to [see] how well pose estimation could perform on the Raspberry Pi. Google provides code to run pose estimation on Android and iOS devices — but I wanted to write Python code to interface with and test the model on the Pi."
Dell's project runs on a Raspberry Pi 4GB, the middle-model in a family which extends from an entry-level 2GB - following the launch 1GB model's semi-retirement — to a top-end 8GB, with the Raspberry Pi Camera Module and a small breadboard with an LED, current-limiting resistor, and push button.
The actual work of estimating the pose of users within the camera's field of vision is handled by Google's TensorFlow Lite PoseNet model: When the button, connected to the Raspberry Pi's general-purpose input/output (GPIO) header, is pushed still frames are captured, analysed, and saved with an skeleton overlaid showing how the model tracked the user's motion and pose.