Objective of the PoC

This project applies human pose detection and machine learning to help kids practise Fundamental Movement Skills (FMS). FMS is a set of structured movements (like vertical jump). Proficiency in these fundamental movements will help kids improve athletic performance in a wide range of activities.

In this PoC, we will use machine learning to detect frames with the kid in 2 key states in a vertical jump - squat and jump peak.

The next step will be to evaluate the kids performance using standard FMS criteria for squat and jump peak.

Model Training:

Trained model using 2000 images of vertical jumps. Images were tagged as squat, jump peak or transition (anything in between).

Result:

Model has 70% accuracy on test set. Some samples below. The mlde detects the squat and jump_peak frame.

Running the Pipelines

This repo has the following key Pipelines:

• pipeline_critical_pose.py
• pipeline_human_pose_points.py
• pipeline-vjump-report.py

pipeline_critical_pose.py

EC2 Setup

Virtual Machine Hardware Specifications 1 Ubuntu 18.04 2 HDD 20 GB

Software Install Guide

Once in the AWS virtual machine, do the following steps

Install Conda

* wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
* chmod +x Miniconda3-latest-Linux-x86_64.sh
./Miniconda3-latest-Linux-x86_64.sh
* rm ./Miniconda3-latest-Linux-x86_64.sh

Create and activate python environment

* conda create -n fastai-dev python=3.6
* conda activate fastai-dev

Install packages:

cv2, scipy, boto3,shutil, pandas pip install <

Git clone the w_yrs_poc1 repo

AWS Configuration

aws configure

Running the Code

python3 pipeline_critical_pose.py --file=<<filename_in_s3_bkt_w-yrs-input-video>> --rotate <<angle>> -o -d