The outputs of the human pose estimation are the x and y locations (on an RGB image) of the user's shoulders, waist, and hip locations.
These x and y coordinates are sent to Max MSP, a visual programming language, where I was able to calculate the distance between those coordinates.
On the right side of the video below you can see the distance (in pixels) calculated between joint locations in real time.
Users can then be identified as one of the five body shapes based on the proportion of the distance between their shoulder, waist, and hip width.