ai-posture-monitor 0.0.22

A package designed to predict static pose and detect falls with 2D RGB Camera in well lit indoor environments.

AI Posture Monitor

This project introduces an innovative, cost-effective solution for real-time activity monitoring of elderly individuals. By leveraging the MediaPipe pose estimation model, fuzzy logic, and finite state machines, the system can reliably track individuals, recognize static postures (standing, sitting, lying), and detect transitions, particularly focusing on falls.

Source code: https://github.com/pat2echo/AI-Posture-Monitor
Dataset (CC BY 4.0): https://www.kaggle.com/datasets/patrickogbuitepu/posture-monitor-and-fall-detection
Demo notebook on Kaggle: https://www.kaggle.com/code/patrickogbuitepu/fall-detection-posture-classification-starter

Requirements

Python 3.9 - 3.12 (mediapipe, the pose-estimation dependency, ships no wheels for 3.13+ in the supported version range).

Supported environments: local CLI/Jupyter and Kaggle notebooks. Google Colab is not officially supported (its runtime preloads libraries that conflict with mediapipe's pinned dependencies).

Quickstart

Set up a virtual environment and install:

# Linux / macOS / WSL (use python3.12/3.11/3.10 if your default python3 is 3.13+):
python3.12 -m venv .venv && source .venv/bin/activate
pip install ai-posture-monitor kagglehub

# Windows:
py -3.12 -m venv .venv && .venv\Scripts\activate
pip install ai-posture-monitor kagglehub

Classify a static pose in a few lines (fetches a sample image from the public Kaggle dataset - no Kaggle account needed):

import kagglehub
import ai_posture_monitor as pm

img = kagglehub.dataset_download(
    'patrickogbuitepu/posture-monitor-and-fall-detection', path='train/pose/stand.jpg')

pose, mp_drawing, mp_pose = pm.initialize_mediapipe()
results, img_rgb, landmarks = pm.detect_pose_landmarks(img, pose=pose, show=True)
print(pm.get_features(landmarks[['X', 'Y', 'Z']].to_numpy(), image_name='stand.jpg'))
# -> ['stand.jpg', True, 59, 'standing', 'standing', 82, 86, 'non_sitting', ...]

CLI demo

A ready-made command-line demo (prints a stand/sit/lie verdict table and saves landmark-annotated images) lives in the repository:

git clone https://github.com/pat2echo/AI-Posture-Monitor
python AI-Posture-Monitor/examples/cli_demo.py                  # 3 sample images from the dataset
python AI-Posture-Monitor/examples/cli_demo.py --images my.jpg  # your own photos

Try it on Kaggle (zero setup)

Open the demo notebook and click "Copy & Edit": https://www.kaggle.com/code/patrickogbuitepu/fall-detection-posture-classification-starter

Fall detection on video

import ai_posture_monitor as pm

pe = pm.PoseEstimation()
pe.process_video(video_file='my_video.mp4', plot_results=True)                      # fall detection
pe.process_video(video_file='my_video.mp4', plot_results=True, predict_fall=False)  # posture only

Features

• Real-Time Activity Monitoring: Tracks and analyzes movements continuously.
• Pose Detection: Identifies key postures: standing, sitting, and lying down.
• Fall Detection: Detects falls with minimal false alarms.
• Fuzzy Logic Analysis: Utilizes fuzzy logic for accurate movement interpretation.
• User-Centric Design: Tailored for elderly individuals living alone.
• Environmental Adaptability: Functions effectively in well-lit indoor settings.
• Scalable and Cost-Effective: Affordable solution for diverse applications.