A Deep Learningaugmented Smart Mirror To Enhance Fitness Training

This Smart Mirror Puts A Personal Trainer In Your Reflection Youtube

In recent years, computer engineers and scientists have developed a number of technological tools that can enhance their fitness training experience; These include smartwatches, fitness trackers, sweat-resistant headphones, smart home gym equipment and smartphone apps. Newly sophisticated computational models, especially deep learning algorithms, can further improve these tools to better meet the needs of individual users.

Scientists at Italy's University of Brescia recently developed a smart mirror computer vision system that can improve exercise at home and in the gym. The system, presented in an article published by the International Association of Sports Biomechanics, is based on a deep learning algorithm trained to recognize human gestures in videos.

"Our business partner ABHorizon has developed a product concept that can guide and educate you in your personal fitness training," Bernardo Lanza, one of the researchers leading the study, told Tech Explorer. "This device shows you the best way to exercise according to your specific needs. To further develop this device, you asked us to investigate the effectiveness of integrated visual system exercise assessment."

A low-cost computer vision system developed by Lanza and his colleagues can use a skeleton algorithm (ie, a deep learning algorithm can extract skeletons from images) running on a Nvidia Jetson Nano internal device with two fisheye cameras. You train the system to run and detect it.

"The vision system we developed can extract information from images using artificial intelligence algorithms," Lanza said. "Our recent work demonstrates the accuracy of our system for measuring arm movements in simple exercises such as bicep curls."

In a previous study, the researchers presented a software design that could be used to create a general prototype of a smart fitness mirror envisioned by the EU-Horizon. Their goal was to create a device with low manufacturing costs, high efficiency and low energy consumption.

"The main advantage of our system is that nothing is connected to the user," Lanza explained. "Through cameras and artificial intelligence applications, we understand and evaluate body movements, recognize posture errors and analyze overall physical activity. Currently, our system's analysis is based on simple body variables (corner angle, hand position...), but we are improving the estimation. The full potential of the car We are working to make it happen.

The smart mirror that Lanza and his colleagues have developed can provide fitness exercises that rank with personal trainers or better. For example, it allows users to set repetitions for specific exercises, as well as define core movements for different body parts (such as stretches, twists, swings, etc.).

All fitness-related data is captured and calculated by the watch in real-time, so users can track it during training or use it to improve exercise results. Lanza and colleagues evaluated the computer vision system in a series of tests, focusing on its ability to predict fitness and track users' bipedal activity.

"We evaluated the accuracy of the visual system to understand different levels of learning," Lanza said. "In traditional biomechanical analysis, the exact accuracy of our measurements is not accepted, but we analyze the whole body's dynamics. This method allows us to identify and understand physical movements and their characteristics."

The researchers found that with well-designed and optimized software, their low-cost vision system can provide valuable fitness information while users perform simple physical activities. Combined with a smart mirror developed by AB-Horizon, the new system will greatly help users exercise at home or at the gym without the supervision of a trainer.

So far, Lanza and his colleagues have evaluated the performance of their system. But now they're developing a prototype that displays the results of their system's analytics on a smart mirror built into a motorized exercise machine.

"We partnered with our business partner AB-Horizon for this project," Lanza said.

The prediction Lanza and colleagues are working on should be able to interpret qualitative data by analyzing raw physical activity data. To train this model, the researchers first collect a large amount of data through fitness tests with athletes and experienced fitness trainers.

Additional information: Bernardo Lanza, Cristina Nuzzi, Simone Pacinetti, Matteo Lancini, Deep Learning for Jim Gesture with Vision-Based Realistic Smart Mirrors, 40th International Society of Biomechanics in Sport, Liverpool, United Kingdom; 19-22 July total. .edu/isbs/vol40/iss1/87/

Quote: Smart glasses powered by deep learning to improve fitness training (September 13, 2022). Retrieved September 14, 2022, from https://techxplore.com/news/2022-09-deep-learning-augmented-smart-mirror. HTML:

This document is protected by copyright. No part may be reproduced without written permission except for personal study. The content is provided for informational purposes only.