Fall Detection and Posture Classification Dataset

Self-recorded video and static-image dataset supporting the dissertation "AI-Driven Posture Analysis Fall Detection System for the Elderly" (University of Essex). The system uses MediaPipe pose estimation, fuzzy logic, and a finite state machine to classify static postures (standing, sitting, lying, bending, squatting, etc.) and detect transitions - particularly falls - from monocular RGB video. Subject and consent: all images and videos feature a single consenting adult (the researcher) performing self-recorded daily activities and posture/fall scenarios. No other individuals appear in the data. Public release is covered by signed self-consent under University of Essex research ethics guidelines. Contents (in posture-monitor-and-fall-detection.zip, structure preserved): train/ - 3 training videos (~11 min total) with per-second activity/fall label CSVs valid/ - 7 validation videos (~26 min total) with matching labels train/pose/ - 113 static pose images (standing, sitting, lying, bending, squatting, falling, occlusion and edge cases) plus MediaPipe-derived feature CSVs and rule-based classifier outputs train/pose-1st-iteration/ - earlier capture/processing pass of the same poses, kept for provenance train/results of static pose classifier on training videos/ - per-frame classifier output for the training videos Accompanying files: README.md (dataset card), column_descriptions.md (schema of every CSV), file_descriptions.md (per-file descriptions), CITATION.cff, LICENSE.txt. Related resources: the dataset is also published on Kaggle; the accompanying software is on GitHub and PyPI (ai-posture-monitor), with a runnable demo notebook on Kaggle.