Unit

Body measurements guidance system using computer vision and pose estimation

Solution

The solution built is a Python Flask application that processes images of through a custom mobilenet model that is adpated from the Pose Net model.

Architecture

Technologies

• Python Flask
• Tensorflow
• Anaconda (Windows)
• Bluetooth serial (Arduino BluNo)
• Html/CSS/JS

System Requirements

• Python >= 3.6.5
• GPU (Optional)
• PIP installed

Installation

1. Install Anaconda from here (Anaconda >= 5.3)

2. Install Swig from here here

3. Open Anaconda prompt:

   Go with the mouse to the Windows Icon (lower left) and start typing "Anaconda". There should show up some matching entries. Select "Anaconda Prompt".

4. Clone our repository and change directory:

   $ git clone https://github.com/Sri-vatsa/Unit
   $ cd Unit
   

5. Create & Activate virtual env:

   $ conda create --prefix=unit python=3.6.5
   

   Note: allow conda to install all relevant packages for python 3.6.5 by typing y when prompted

   List all available conda environments: (Optional)

   $ conda info --envs
   

   Activate conda environment for Unit:

   $ conda activate PATH/TO/ENV
   

   e.g. PATH/TO/ENV : D:\Unit\unit

6. Install dependencies:

   $ pip install -r requirements.txt
   

7. Make paf for pose estimation post processing:

   a. Change directory to pafprocess:

   $ cd tf_pose_estimation/tf_pose/pafprocess
   

   b. Build the pafprocess module

   $ swig -python -c++ pafprocess.i && python3 setup.py build_ext --inplace
   

8. change directory back to root directory:

   $ cd ../../..
   

Run application

1. Run backend:

$ python main.py

2. Find your local ip address:

In the anaconda prompt, type:

$ ipconfig 

IPv4 Address is your local ip address.Take note of this address.

E.g. 172.17.10.153

3. Run the product demo in a browser

a. Open any browser (Chrome is preferred) b. In the addressbar, type : https://YOUR_IP_ADDRESS:5000 E.g. https://172.17.10.153:5000

Note: If the browser alerts that the site is insecure, click on advanced settings -> continue anyway