Multi Person Pose Estimation

This blog is about the new model that Google has released called the Movenet model for multi person pose estimation.

This model is built on top of these 3 papers:

• MobileNetv2: https://arxiv.org/pdf/1801.04381.pdf - backbone network
• Feature Pyramid Networks: https://arxiv.org/pdf/1612.03144.pdf - feature extraction
• CenterNet: https://arxiv.org/pdf/1904.07850.pdf - prediction heads - the layer that outputs the key points.

Steps for the project

1. Install Multi Person Movenet : Tensorflow and Tensorflow Hub
2. Make Detections using live video feed and webcam
3. Draw key points and edges

How it works

1. Install Tensorflow and Tensorflow Hub
2. Load the model
3. Make detections from videos, camera and render.