AI-Based Health Risk Prediction Using Real-Time Pose Estimation and Deep Learning

Abstract

Preventive healthcare needs early health risk detection of conditions like obesity, malnutrition, and metabolic disorders. Conventional screening procedures are either expensive, demand trained staff, and physical visits, hence less accessible and scalable. This paper introduces a real-time, AI-powered health risk prediction system that employs computer vision and deep learning to gauge physical health with a webcam alone. The system proposed utilizes MediaPipe and OpenCV to infer body pose landmarks from live video feed. Anthropometric features like height, BMI, shoulder width, arm length, and leg length, as well as waist-to-hip ratio, are automatically computed from these. These values, together with user-supplied information such as age and gender, are input into a deep learning model that has been trained to predict an individual as being in one of four categories: healthy, risk of obesity, malnourishment, or risk of metabolic disorder. All processing is carried out in real time and aggregated into a web-based interface, providing instant feedback, graphical insights, and health suggestions to the user. The backend is implemented with Python, TensorFlow/Keras, and Flask for rapid processing and model inference. This system illustrates the promise of affordable, non-invasive technologies for initial health testing. It can be used in telemedicine, fitness programs, and distant health screening, increasing preventive care to be more scalable and affordable.

Introduction

The global rise in non-communicable diseases (NCDs) like obesity and malnutrition challenges public health, especially in under-resourced areas where traditional assessments require costly tools and expertise. Advances in computer vision and AI, particularly human pose estimation (HPE) via webcams and smartphones, offer a low-cost, accessible alternative by extracting body measurements from video to assess health risks.

This research proposes a web-based system that combines pose estimation (using tools like MediaPipe) with deep learning models to predict health risks such as obesity, malnutrition, metabolic disorders, or healthy status in real time. The system analyzes skeletal landmarks to calculate anthropometric features (height, limb lengths, waist-to-hip ratio) and integrates user demographic data for accurate health classification.

The deep neural network model, trained on over 2,000 health records, achieves high accuracy (~92.4%) in categorizing physical health threats. The platform includes a user-friendly web interface for real-time posture capture, visualization of results, and optional email reports, making it suitable for telemedicine and preventive health screening without specialized equipment.

Overall, this approach demonstrates how AI-driven, vision-based health monitoring can increase accessibility, reduce costs, and provide personalized early health risk detection.

Conclusion

This work presented a real-time AI-powered health risk assessment system integrating pose estimation, anthropometry measurement extraction, and deep learning-based classification into a single non-invasive framework. With the use of common webcams and clean frontend design, the system allows for easy screening of health risks like obesity, malnourishment, and metabolic syndrome with minimal effort. Model evaluation showed excellent classification accuracy (92.4%) and stable performance across test samples, validating the applicability of computer vision in preventive healthcare. The combination of a dynamic frontend, backend inference automation, and email reporting as an option provides end-to-end user experience—from pose tracking to actionable knowledge. Next steps include • Increasing the dataset size to cover more varied body types and demographics • Adding explainable AI to develop user trust • Integration with electronic health records and third-party fitness APIs This system is a significant step towards affordable, scalable, and smart health screening. It presents possible use cases in telemedicine, wellness apps, and population-scale community health programs.

References

This work presented a real-time AI-powered health risk assessment system integrating pose estimation, anthropometry measurement extraction, and deep learning-based classification into a single non-invasive framework. With the use of common webcams and clean frontend design, the system allows for easy screening of health risks like obesity, malnourishment, and metabolic syndrome with minimal effort. Model evaluation showed excellent classification accuracy (92.4%) and stable performance across test samples, validating the applicability of computer vision in preventive healthcare. The combination of a dynamic frontend, backend inference automation, and email reporting as an option provides end-to-end user experience—from pose tracking to actionable knowledge. Next steps include • Increasing the dataset size to cover more varied body types and demographics • Adding explainable AI to develop user trust • Integration with electronic health records and third-party fitness APIs This system is a significant step towards affordable, scalable, and smart health screening. It presents possible use cases in telemedicine, wellness apps, and population-scale community health programs.

Copyright

Copyright © 2025 Suhas Revansiddha Amane. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.