Datasets

Below is a curated list of datasets developed by the Cyber Identity & Behavior Research (CIBeR) Lab at the University of South Florida.

🗂️ Available Datasets

GestDoor: IMU-Based Door Entry Biometric Dataset

Description:
GestDoor contains wearable sensor data collected during door-opening interactions to support research in motion-based authentication, behavioral biometrics, and gesture recognition. Using two 6-DOF IMUs (wrist + upper arm), 11 participants performed four task types across up to three sessions, producing 3,330 segmented samples of accelerometer and gyroscope data sampled at 100 Hz.

Includes:

• 6-axis acceleration + angular velocity (+ quaternions, timestamps)
• Four door-opening labels: L_PUSH, L_PULL, R_PUSH, R_PULL
• Participant/session metadata (age, sex, height, dominant hand, session count)
• Fully segmented samples — no preprocessing required

Suggested Uses:

• Motion-based authentication
• Smartwatch and wearable security research
• Gesture and activity recognition
• Biometric permanence and cross-session analysis
• Behavioral signal modeling
• Sensor fusion experimentation

Usage Notes:

• Files provided in .csv format
• Load using Python (numpy, pandas) or MATLAB
• Suitable for ML models (SVM, RF, KNN), signal-distance approaches (DTW), or feature pipelines
• Supports intra-session and cross-session evaluation protocols

Citation:

M. Ebraheem and T. Neal, “GestDoor: Gesture-Based User Authentication for Door Entries Utilizing Wearable IMUs,” 2025 IEEE 19th International Conference on Automatic Face and Gesture Recognition (FG), Tampa/Clearwater, FL, USA, 2025, pp. 1-8, doi: 10.1109/FG61629.2025.11099107.

CD3: Cross-Domain Deception Dataset

Description:
The Cross-Domain Deception Dataset (CD3) contains frame-level visual features extracted from interview video recordings to support research in deception detection through facial expressions, action units, gaze, and body/hand gestures. Using a commercial laptop and Microsoft Teams, 45 participants completed mock interviews across two sessions, responding to questions about biography, academic success, and well-being.
The dataset provides 1,270 truthful and 587 deceptive clips, enabling cross-domain analysis of how deception appears differently across content areas and supporting research into well-being–specific deception models.

Includes (983 frame-level features per sample):

• Gaze: 8 gaze features (direction vectors + angles)
• Landmarks:
  • 136 × 2D facial landmarks
  • 204 × 3D facial landmarks
  • 112 × 2D eye landmarks
  • 168 × 3D eye landmarks
  • 140 × 2D face keypoints
• Head Pose: translation (x,y,z), rotation (pitch, yaw, roll)
• Face Shape: 40 PCA-based shape parameters
• Facial Action Units: 35 total (18 presence, 17 intensity)
• Body & Hands:
  • 50 body keypoints (2D)
  • 84 hand keypoints (2D)
• Labels & Identifiers: deception label (1 = deceptive, 0 = truthful), participant ID (PXXX)

File Format:

• Provided in .csv format
• Each row represents one video frame from a participant response

Suggested Uses:

• Deception detection (cross-domain + well-being–specific)
• Action unit and gaze–based behavioral modeling
• Gesture and micro-expression analysis
• Domain adaptation and cross-domain inference
• Multimodal vision features for cognitive state estimation
• Representation learning for social/behavioral computing

Educational & Research Use:
Available for coursework, capstone projects, theses, and experimentation in deception detection, behavioral modeling, and multimodal machine learning.

Citation:

S. L. King and T. Neal, “Exploring Vision-Based Features for Detecting Deception in Well-Being: A Cross-Domain Comparison,” 2025 IEEE 19th International Conference on Automatic Face and Gesture Recognition (FG), Tampa/Clearwater, FL, USA, 2025, pp. 1-10, doi: 10.1109/FG61629.2025.11099290.

📥 Requesting Access or Citing

You may contact Dr. Tempestt Neal for dataset access, as necessary, or collaboration inquiries.
When using CIBeR datasets, please include the appropriate citations.