“The only failure is not to try, because your effort in itself is a success.” - Sister Madona Buder aka Iron Nun

The COVID-19 pandemic had increased loneliness, a sense of despair and the digital divide among older generations who lacked computer literacy. Elderly citizens across the globe felt helpless as well as became physically inactive with the daily curfew regulations and home isolation rules. Despite knowing the benefits of physical activities, they fear body injuries, lack confidence and were afraid to exercise without the presence of a trainer. In addition, they lack technological prowess; unable to browse online search engines about body training regimes.

'Exercise Program for the Elderly with ML Tech' is one responsive measure by two design researchers amid the pandemic which utilises Tensorflow.js with PoseNet to encourage and guide senior citizens to exercise independently from their homes along with fun music. This experimental academia project provides easy poses for the users to follow, it focuses on correcting postures (pose-estimation algorithms) of the body and playing a piece of energising music to encourage the users. As soon as the user matches with the given poses, the application proceeds to present the next pose. It is easy, free of cost, convenient and can be performed remotely from one’s home. This idea was fruitful during the 2019 pandemic time when it was difficult for senior citizens to move out of their homes and access fitness centers.

View the project outcome

The booming use of Artificial Intelligence technology in our daily lives lies in their innovation, accessibility, machine learning algorithms and ease of natural language processing.

Why PoseNet was chosen? PoseNet is a pre-trained pose recognition tool which is helpful in pattern identification with minimal human effort. Tensorflow.js was used to curate a customisable workout plan. This machine learning application was integrated to help senior citizens execute and follow the correct postures for exercising.

No language barriers - A universal medium was needed which could mimic human working out abilities, correct postures in real-time and provide visual guidance instead of conventional textual guides for ageing users.

Realtime Posture Correction - Tensorflow.js with PoseNet facilitates pose estimation and recognition which is helpful in pattern identification with minimal human effort. This application was needed to help senior citizens to practise and follow some correct postures for exercising.

Accessible - This online system can be accessed anywhere and anytime across the globe. Amid situations like COVID-19, it is convenient, safe and easy for older adults (60+ years old) to exercise remotely and have a virtual trainer rather than going out to a gymnasium or fitness centre.

Customise the experience and exercises - Nowadays numerous tutorials and journals are available on online platforms for any non-coders (people lacking computer programming skills) to learn Machine Learning software, especially for writers, futurists, designers and musicians etc. This will aid anyone to upgrade and customise the application (Exercise Program for the Elderly with ML Technology). Furthermore, they can turn the system into a personalised platform to cater to different needs since every individual's requirements are unique.

Fig 1 - Training the Model

Fig 2 - Exploring a few exercise poses

What is ml5.js?

Ml5.js is an open-source AI library and web-friendly ML project that spun off from TensorFlow.js in collaboration with NYU's Interactive Telecommunications/Interactive Media Arts program school. Ml5.js utilizes the local Graphics Processing Units (GPUs) of users through web browsers. The purpose of developing ml5.js is to make more ML libraries accessible and easier to use for the mass (students, researchers, creative etc). Click here to learn more.

Research & Inspiration -

AI Aerobics - To have some fun & match your posesClick here

Created by SPACE10, AI Aerobics is a human-computer interaction (HCI) platform where the computer will instruct the human/user to follow a series of movements. The poses you’ll be asked to demonstrate have been generated by a reinforcement learning algorithm in a simulated environment. That means our machine has learned how to move by following its own logic and rules—isolated from practically any human influence or bias (unsupervised learning).

Creators -

Bomi Do & Rittika Basu

OCAD University, Toronto

https://tinyurl.com/47k4n9yt