IPPT Anywherez

• GIF

Inspiration

We were moved by the observation that lower income NSmen do not have the opportunity to take as many IPPT attempts in SAF Camps (i.e Maju or Kranji Camp) simply because they do not have the luxury of time. Nor they have the luxury of money to book expensive appointments at True Fitness to gain access to their proprietary motion sensor machines. Additionally, it is very difficult for each NSMen to check on each other's pushups/situps form as everybody is different and has different standards for themselves.

Therefore as an act of service to give back to the nation to symbolize our gratitude for making us serve 2 years of National Service, we figured out the secret recipe behind the seemingly-cryptic IPPT machines and decided to democratize access to ELISS machines by making them fully open-source. Now that it's publicly available and accessible, everyone (NSFs, NSMen and Women) will be able to train and check their form to maximize training efficiency and so that they can earn greater incentives when clearing IPPT each year.

These IPPT incentives in cash can be used to defray costs of living, hence ameliorating their status quo, potentially lifting them out of penury.

What it does

This innovative system harnesses the power of artificial intelligence and computer vision technology to track the number of repetitions and ensure proper form when executing the IPPT exercises of Pushups and Situps. Users can also playback the video to see their own form from a third party's perspective, enabling them to personally identity further rooms for improvement.

How we built it

• OpenCV to enable the visual rendering of webcam video captures
• MediaPipe to identify and draw the pose estimation to predict and track the person's key joints (hip, elbow, shoulder etc.)
• Trigonometry to calculate the angles between key joints to identify and evaluate whether a repetition is valid
• Streamlit to facilitate the deployment of the OpenCV model which is hosted on Microsoft Azure with Cloudflare protection
• GitHub for version control

Challenges we ran into

Initially, a lot of the sit ups that we executed went undetected by our program code. (interestingly enough, the actual ELISS machines are plagued by the same issue.) We went back to the drawing board and we managed to trace the source of the error by trial and error. In the end, we spent a lot of time to trying to optimize the model for sit-ups detection by playing around with the trigonometry values/formulas between different pairs of joints until we finally got a system that was sufficiently accurate and precise.

Accomplishments that we're proud of

The idea of replicating the IPPT motion sensor machine initially seemed like a Herculean task. Despite certain challenges (integrating the webcam video capture with pose estimation, detection problems etc. ) when building the model initially, we are proud that we managed to produce a working, fully-functional, marketable prototype within the span of 24 hours.

What we learned

We learnt that with advances in processing power, AI/ML workloads can be executed anywhere, making IPPT machines lower cost.

What's next for IPPT Anywherez

• Include a timer and track the user's progress consistently over a month-long, or year-long period, motivating users to maintain their physical fitness for a protracted period of time
• Vertical integration with other fitness services, such as FitBit or Apple, so that this computer vision technology can be integrated into their fitness regime, enabling users to maximize their training efficiency.
• Sourcing open-source pushups/situps data and training a neural network architecture that allows our OpenCV model to adapt to male/female, account for different body sizes and age, reducing the model's bias for the younger audience.

Built With

• azure
• flask
• machine-learning
• math
• opencv
• python
• streamlit