Enhancing Upper Limb Rehabilitation with Unsupervised Machine Learning Models
summary
A pipeline that ingests eight joint-angle streams from a Unity AR/VR puzzle game, extracts median and variance into 16 features, applies PCA to 3 components to discard noise, then clusters with K‑means → 3 clusters and MiniSom to label movement phenotypes; a Panel dashboard visualizes biomechanical metrics for clinicians. Implementation uses Python, NumPy, Pandas, scikit‑learn, MiniSom, deployed in containers as a browser-accessible SaaS; works on laptop webcams but suffers webcam noise ~5–10° limiting fine-grained measures.
AI-written summary · about
chapters (11)
- 0:00Stroke rehab research intro
- 1:52VR/AR rehab games design
- 3:40Data pipeline for unstructured data
- 6:09Raw movement to features
- 9:01PCA for noise reduction
- 11:06K-means clustering labels
- 15:25Physiotherapist cluster validation
- 17:34Dashboards for healthcare pros
- 20:25Outliers in stroke data
- 21:32Future PhD research
- 23:58Q&A on setup & collection
related talks
17mLearning to assess the quality of stroke rehabilitation exercises
47mMachine Learning for medical rehabilitation
50mBuilding Intelligent Mobile Apps using Machine Learning
22mSeeing, Sensing and Recognizing Laban Movement Qualities
28mDeep Learning-Based Human Activity Recognition for Physical Skills...
44mFrom Athlete to Algorithm: Transforming Canoe Technique Analysis with AI
06mUnsupervised Human Activity Representation Learning with Multi-task Deep Clustering
27mAnalysis of sports scenes using computer vision models