MPSC-RR Dataset

Name: MPSC-RR
Creator: Snehalraj Chugh
License: https://choosealicense.com/licenses/cc-by-4.0/

Respiratory Waveform Reconstruction using Persistent Independent Particles Tracking from Video

Dataset Description

The MPSC-RR dataset is a comprehensive collection of RGB videos designed for contactless respiratory rate estimation and respiratory waveform reconstruction research. The dataset captures respiratory-induced movements across multiple body positions and camera angles, providing diverse scenarios for developing robust respiratory monitoring algorithms.

Key Features

Multi-Position Coverage: Front, back, side, and lying positions
Natural Breathing Patterns: Regular, slow, and fast breathing captured
Diverse Demographics: Adult volunteers across different age groups and backgrounds
Ground Truth Annotations: Vernier Go Direct Respiration Belt measurements
High-Quality Videos: RGB videos with clear respiratory motion visibility
Real-World Conditions: Varied lighting, clothing, and environmental settings

Dataset Statistics

Total Videos: 31 video sessions
Subjects: 24 adult volunteers (Subject IDs: 1-24)
Duration: 2.5-5 minutes per video (Average: ~4 minutes)
Resolution: Varied resolutions from 304×304 to 1602×1080
Frame Rate: Primarily 30 FPS (with some at 30.1, 29, and 24 FPS)
Total Duration: ~2.1 hours of respiratory monitoring data
Total Size: 5.18 GB

Data Structure

Video Files

s{subject_id}_{view_position}.{ext}

Naming Convention:

s1_front.mov - Subject 1, front view
s20_side.mp4 - Subject 20, side view (Note: one video in MP4 format)
s24_lying.mov - Subject 24, lying position

View Positions:

front - Frontal chest/abdomen view (6 videos)
back - Back/shoulder movement view (4 videos)
side - Side profile view (16 videos)
lying - Lying down position (5 videos)

Metadata Structure

{
    "video_id": str,                    # e.g., "s1_front"
    "subject_id": int,                  # Subject identifier (1-24)
    "view_position": str,               # front, back, side, lying
    "duration_seconds": float,          # Video duration
    "resolution": str,                  # e.g., "1920x1080"
    "frame_rate": float,                # 30.0, 30.1, 29.0, or 24.0 FPS
    "file_size_mb": float,              # File size in megabytes
    "filename": str                     # Actual filename with extension
}

Usage

Loading the Dataset

from datasets import load_dataset

# Load all videos
dataset = load_dataset("justchugh/MPSC-RR")

# Access video information
sample = dataset[0]
video_id = sample["video_id"]
subject_id = sample["subject_id"] 
view_position = sample["view_position"]

Filtering Videos

# Get front view videos
front_videos = dataset.filter(lambda x: x["view_position"] == "front")

# Get specific subject's videos
subject_1 = dataset.filter(lambda x: x["subject_id"] == 1)

# Get lying position videos
lying_videos = dataset.filter(lambda x: x["view_position"] == "lying")

Basic Analysis

# Count videos per position
positions = [sample["view_position"] for sample in dataset]
position_counts = {pos: positions.count(pos) for pos in set(positions)}
print("Videos per position:", position_counts)

# List all subjects
subjects = set(sample["subject_id"] for sample in dataset)
print(f"Subjects: {sorted(subjects)}")

Data Collection

Equipment

Camera: Standard RGB cameras (mobile phones, mounted cameras)
Ground Truth: Vernier Go Direct Respiration Belt for pressure measurements
Distance: 1-1.5 meters from subjects
Environment: Clinical laboratory setting with controlled conditions

Protocol

Subject Preparation: Comfortable positioning with clear view of respiratory regions
Baseline Recording: 30-second calibration period
Data Collection: 3-5 minutes of natural breathing
Ground Truth Sync: Synchronized pressure belt data collection
Quality Check: Manual verification of respiratory cycles

View Positions Explained

Position	Description	Captured Regions	Use Case	Count
`front`	Frontal view	Chest, abdomen expansion	Standard respiratory monitoring	6
`back`	Back view	Shoulder, back movement	Posterior respiratory motion	4
`side`	Side profile	Chest wall movement	Lateral respiratory dynamics	16
`lying`	Supine position	Abdomen, chest (lying)	Sleep/rest respiratory patterns	5

Dataset Distribution

Subject Range	Video Count	Positions Available
1-2	6 videos	front, back, lying
3	2 videos	back, lying
4	1 video	front
5	2 videos	front, lying
6	1 video	front
7-22	16 videos	side (primarily)
23	1 video	front
24	2 videos	back, lying

Benchmark Results

Performance of state-of-the-art methods on MPSC-RR:

Method	MAE (bpm)	RMSE (bpm)	Modality
Intensity-based	3.25	5.12	RGB
Optical Flow	2.42	3.89	RGB
PIPs++	1.62	2.92	RGB
RRPIPS	1.01	1.80	RGB

Applications

This dataset supports research in:

Contactless Vital Sign Monitoring
Multi-Position Respiratory Analysis
Computer Vision for Healthcare
Sleep and Rest Monitoring
Wearable-Free Health Tracking
Clinical Decision Support Systems

Data Quality

Inclusion Criteria

Clear visibility of respiratory-induced motion
Stable video recording (minimal camera movement)
Synchronized ground truth data available
Adequate lighting conditions

Quality Metrics

Motion Clarity: All videos show visible respiratory movement
Synchronization: <50ms offset between video and pressure data
Duration: Minimum 148 seconds per recording
Resolution: Varied resolutions optimized for respiratory motion capture

Related Datasets

AIR-125: Infant respiratory monitoring (125 videos, infants)
BIDMC: PPG and respiration signals (53 recordings, clinical)
Sleep Database: NIR/IR respiratory data (28 videos, adults)

Ethical Considerations

IRB Approval: All data collection approved by institutional review board
Informed Consent: Written consent obtained from all participants
Privacy Protection: Faces blurred or cropped when necessary
Data Anonymization: No personally identifiable information included
Voluntary Participation: Participants could withdraw at any time

Citation

If you use this dataset in your research, please cite:

@inproceedings{hasan2025rrpips,
    title={RRPIPS: Respiratory Waveform Reconstruction using Persistent Independent Particles Tracking from Video},
    author={Hasan, Zahid and Ahmed, Masud and Sakib, Shadman and Chugh, Snehalraj and Khan, Md Azim and Faridee, Abu Zaher MD and Roy, Nirmalya},
    booktitle={ACM/IEEE International Conference on Connected Health: Applications, Systems and Engineering Technologies (CHASE)},
    year={2025},
    pages={1--12},
    doi={10.1145/3721201.3721366}
}

Quick Start

# Install required packages
pip install datasets

# Load and explore dataset
from datasets import load_dataset
dataset = load_dataset("justchugh/MPSC-RR")

print(f"Total videos: {len(dataset)}")
print(f"First video: {dataset[0]['video_id']}")
print(f"Positions available: {set(s['view_position'] for s in dataset)}")

# Example: Get all side view videos
side_videos = dataset.filter(lambda x: x["view_position"] == "side")
print(f"Side view videos: {len(side_videos)}")