AI Optimizes School Fitness Evaluations

By Dr Silpaja Chandrasekar, PhDReviewed by Susha Cheriyedath, M.Sc.Aug 27 2024

In a paper published in the journal Applied Sciences, researchers developed a multiplatform computer vision (CV) system for evaluating schoolchildren's physical fitness using smartphones.

The study involved 228 children and compared traditional evaluation methods with the new system, achieving high accuracy in field and lab tests. The mobile app also demonstrated notable accuracy and user satisfaction, indicating that integrating CV into physical fitness assessments was feasible and effective.

Related Work

Past work has highlighted the rise in sedentary lifestyles among schoolchildren and its negative impact on health, leading to increased obesity rates. Traditional physical fitness assessments have been criticized for their subjectivity and reliance on the assessor's experience. Recent CV and artificial intelligence (AI) advances have provided non-intrusive, accurate methods for monitoring and evaluating physical activity.

Study Overview Summary

A descriptive cross-sectional study was conducted with 228 schoolchildren (128 boys and 108 girls) aged 8 to 18 from three schools in Arequipa, Peru. The study adhered to the Declaration of Helsinki and received approval from the local ethics committee (UCSM-035-2023).

Participants included those who regularly attended physical education classes and provided consent. Exclusions were made for those with physical injuries or absent on the evaluation day. The study utilized anthropometric measurements, including weight, height, and waist circumference, and employed physical tests such as sit-and-reach, horizontal jump, biceps curl, and abdominal crunches, all performed under standardized conditions.

Analysts conducted the physical tests using techniques and instruments that adhered to established protocols. Anthropometric measurements were recorded using precise instruments, including a seca digital scale, an aluminum stadiometer, and measuring tape.

The physical tests involved a sit-and-reach box for flexibility, a tape measure for horizontal jumps, and a standardized approach for abdominal crunches and biceps curls. Each test was conducted with minimal clothing to ensure accuracy, and the data collected were intended to evaluate flexibility, explosive force, and muscle resistance.

The study integrated CV and AI technologies to address the challenge of subjectivity and data reliability in traditional fitness assessments. The cross-industry standard process for data mining (CRISP-DM) process model guided the implementation, encompassing stages from business understanding to deployment.

The process included traditional data collection, video recording of tests, and analysis using a CV to capture test performance metrics. Testing environments varied from a controlled laboratory using Azure Kinect cameras to field settings with Canon video cameras, with physical education teachers performing validation on mobile devices.

Data preparation involved careful setup to minimize visual noise and ensure accurate joint detection. The LightBuzz was employed to develop the analysis tools for measuring physical tests. Testing in the laboratory and field environments was meticulously planned, with specific camera setups and video capture procedures.

The analysis involved converting pixel measurements to centimeters and applying algorithms to evaluate test performance. Statistical validation was performed using descriptive statistics and Bland-Altman diagrams to compare traditional measurements with software-derived results.

Performance Evaluation Summary

The initial results focused on determining the students' percentiles as a baseline for measuring performance across four physical tests, categorized by age and sex. Percentiles provide a way to gauge a student's performance compared to a reference group. A high percentile (85%) indicates superior performance, while percentiles between the 15th and 85th are deemed adequate, and those below the 15th percentile suggest decreased physical fitness.

The desktop version of the proposed software processes the test videos by selecting the appropriate test scene and configuring the platform to recognize the video type sensor. After importing and processing the video, the software displays the test results, which are then exported to Microsoft Excel 365 for further analysis.

Researchers calculated the error percentage of the software's measurements and assessed the accuracy and precision of data. The results reveal differences in means between traditional and software methods, with broad limits of agreement observed across all four physical tests.

The Systematic Software Quality Model assessed user satisfaction with the prototype, focusing on functionality, usability, and reliability. Teachers evaluated their students' physical fitness using the software and completed a questionnaire based on these categories. The findings showed that the functionalities were 100% satisfied, the usability was 97%, and the dependability was 100%, indicating that the educational users were very satisfied.

Conclusion

To sum up, integrating CV into monitoring physical activity in schoolchildren marked a significant advancement in promoting healthier educational environments. With accuracy ranging from 86.74% to 98.26% across tests, the desktop software showed excellent precision, while the mobile version had an overall accuracy of 88.75%.

Validation through physical education teachers confirmed the system's effectiveness and user satisfaction. However, limitations such as camera quality, lighting, and exercise execution affected accuracy, with the mobile version showing slightly lower precision than the desktop version.