AI-Driven Posture Assessment and Coaching for Personalized Exercise Correction

Use Case:

AI4Pose is a mobile application that evaluates and assists users in performing exercises correctly by analyzing posture and identifying deviations using AI-powered pose estimation. Designed for fitness and rehabilitation contexts it guides users through visual feedback and reports based on machine learning rule generation and geometric analysis.

Outcome:

• Integration of Mediapipe-based pose estimation with machine learning for postural misalignment detection;
• Real-time inference engine with user-specific feedback for exercise execution;
• Custom dataset collected with physiotherapists, including annotated videos, keyframes, and corrective guidelines;
• Lightweight feedback engine using angle gradients and k-means for rule extraction from reference videos;
• Fully functional UI built on Material Design 3 for intuitive user experience across mobile devices;
• Cloud backend (Python server in Docker, Exoscale cloud) ensuring efficient and scalable data processing;
• Addressed variability in camera hardware to minimize distortion effects; Security-by-design architecture aligned with ISO 27001 standards.

Ecosystem Support:

StairwAI enabled access to AI experts for model selection, training strategy, and rule-generation logic. It supported scalability testing, dataset structuring, UI design evaluation, and ensured real-world validation with exercise professionals.

AI Relevance:

This story illustrates how AI4Pose leverages:

1. Markerless, real-time pose estimation for personalized guidance;
2. Human-in-the-loop dataset generation with domain experts;
3. Adaptable architecture for health and fitness applications;
4. Practical AI adoption in mobile wellness tools for broad, non-technical user bases.

Summary:

BioAssist’s AI4Pose is a mobile AI system for real-time exercise evaluation and feedback. It combines pose estimation (Mediapipe) with rule-based feedback mechanisms using vector geometry and k-means clustering to analyze postural accuracy. The solution was developed with input from physiotherapists and validated through user testing. The backend uses a Python-based microservice deployed in the cloud, capable of handling pose analysis, video processing, and personalized feedback generation. The intuitive UI allows users to perform exercises, receive guidance, and understand mistakes through visual analytics. With a clear commercialization roadmap and ISO 27001 compliance, AI4Pose is positioned to enter the digital health and fitness market across Europe.