Move The Next Gen: Motor Skill Data and Scoring Platform
A non-invasive, video-based framework for assessing foundational motor skills in children aged 4-12, with privacy-preserving processing, expert-guided workflows, and AI-assisted scoring for new unseen sessions. Led by Deakin University's Institute for Physical Activity and Nutrition (IPAN).
Project Summary
This work focused on practical deployment: building a lightweight, plug-and-play platform for non-invasive motor-skill assessment in real-world settings such as schools, clubs, and community programs.
The workflow covers foundational movement tasks including hopping, jumping, skipping, throwing, kicking, and catching. After capture, videos are processed through automated anonymization, then routed to expert scoring and downstream model development.
Real-time Visual Action Recognition and Joint Angle Analysis with no Physical Sensor Attached!
My Role
I delivered this project independently as the only machine learning engineer on the team, from initial architecture and foundation-building to deployment-oriented implementation.
- Designed and established a modular end-to-end architecture across capture, anonymization, scoring, and modeling.
- Implemented a lightweight GUI-based recorder and camera capture services for field data collection on phones/tablets.
- Integrated deep-learning anonymization to remove identifiable visual elements before assessment.
- Enabled expert-review workflows and prepared structured labels for supervised model training.
- Trained an end-to-end scoring model for privacy-preserving evaluation of new records.
System Architecture
Technical Pipeline
1) Data Capture
Portable recording interface for consistent, non-invasive video acquisition of foundational motor tasks.
2) Automated Anonymization
Deep model processing removes personally identifiable visual information across frames before scoring.
3) Expert Scoring
Anonymized records are presented for expert assessment in a dedicated, repeatable scoring workflow.
4) Label-Ready Dataset
Processed data is organized for model training with clear links between activities, criteria, and scores.
5) End-to-End Modeling
Final model learns to estimate motor-skill performance from new anonymized recordings with expert-aligned outputs.
6) Deployment Readiness
Framework designed for modular operation, reproducibility, and low-cost rollout without specialist equipment.
Assessment Features
- Non-invasive and video-based assessment of foundational motor competence.
- Real-time processing and feedback to support practical use during sessions.
- Lightweight and plug-and-play framework for accessibility in typical field environments.
- Designed to reduce manual assessment burden while maintaining strong scoring quality.