Body Measurements using CV

Submitted by Ankesh Gupta(2015CS10435), Krunal Shah(2015EE10476), Saket Dingliwal(2015CS10254)

The goal of assignment was to make real-world body part measurements using 2D images. The repository includes methods to measure shoulder distance, wrist-to-shoulder measurement, and waist approximation. For implementation details and other nitty-gritty associated with the project, its recommended to lookup the attached presentation named: Presentation.pdf.

To run code, change to src/ directory and type linux shell:

python code2.py -i1 <path to Image1> -i2 <path to Image2> -i3 <path to Image3> -a <Correction_mode>

Notes on running the code

1. The code has been tested and developed in python2 using Ubuntu 16.04. OpenCV version 2.4.13.6
2. Images required for above code are specific. The details are given below.
3. Correction_mode parameter is the flag, which tells code whether to perform affine + metric correction on the image. Enter True to perform correction else False

Additional Notes

1. First 2 images that pops up is for finding waist circumference. Mark waist ends on both the image.
2. Next 2 images are for calculating shoulder length. Mark fall near neck both sides in first image and mark shoulder, both-sides in second image.
3. Repeat step 2 again. This is again calculating shoulder length for robustness from different image.
4. Mark both the wrist - end of hands nearly in the next popped image.
5. While in 2,3,4 points, there is also projected points shown. You can manually mark or not mark.

Input Image1

This image is with the subject holding a checkered board in hands. This helps measure shoulder distance. Check the image below.

Checkered board is special. Its helps in calibration of camera image world for 3D measurements. If you use any other chess-type board, measure the side length of unit square and change global ref_ht parameter in code2.py.

Input Image2

This image is with the subject spreading out his hands. This helps in wrist-to-shoulder measurement, and provide width of waist's projection.

Input Image3

This image is capturing side-view of subject. This provide thickness of waist and helps complete waist measurement.

Waist is modelled as an ellipse and measured analogous to finding perimeter of ellipse. Hence appromation is mentioned.

When code runs, you will be shown points selected by our heuristic as to-be shoulder/wrist. If suspicious/incorrect, you can explicitly select those points on image and pressing esc key thereafter.