Stereo Egocentric Video Dataset with IMU & Annotations
Egocentric & First-Person Data
Tags and Keywords
$280,000
About
Stereo Egocentric Video Dataset with IMU & Annotations
About
A large-scale collection of approximately 10,000 hours of real-world, head-mounted stereo egocentric video, accompanied by synchronized high-frequency IMU measurements, camera calibration parameters and timestamped natural-language action annotations.
Designed for Physical AI, Embodied AI, Robotics, Vision-Language-Action (VLA) models, imitation learning and multimodal AI development, the dataset captures human activities from a first-person stereo perspective.
Each stereo recording provides two synchronized Full HD views, alongside motion measurements and calibration information that support spatial understanding, motion analysis and learning from human demonstrations.
The dataset is available under non-exclusive commercial licensing, with custom delivery based on buyer requirements.
Data Product Features
1. High-Resolution Stereo Egocentric Video
- Capture Device: Head-mounted stereo camera
- Camera Configuration: Two synchronized RGB views
- Shutter Type: Global shutter
- Per-Camera Resolution: 1920 × 1080 pixels
- Combined Frame Layout: 3840 × 1080 pixels, side by side
- Frame Rate: 30 FPS
- Video Compression: H.265 / HEVC
- Perspective: First-person, human-worn stereo
- Stereo Baseline: Approximately 60 mm
Stereo capture provides two viewpoints of the same activity, enabling depth estimation and spatial interpretation using the supplied calibration information.
2. High-Frequency Inertial Measurements
The recordings include time-aligned 6-axis IMU measurements.
IMU Specifications:
- 3-axis accelerometer
- 3-axis gyroscope
- Sampling frequency of approximately 500 Hz
- Time alignment with stereo video
- Camera-to-IMU calibration parameters
- IMU noise characteristics
These measurements support visual-inertial learning, motion estimation and research into human movement during task execution.
3. Camera & Sensor Calibration
Calibration information includes:
- Individual camera intrinsic parameters
- Lens distortion parameters
- Stereo camera baseline, approximately 60 mm
- Camera-to-IMU extrinsic transformations
- Camera/IMU time offset
- IMU noise profile
The calibration data supports stereo geometry, visual-inertial analysis and sensor-aware AI model development.
4. Timestamped Natural-Language Action Annotations
The dataset includes action-level annotations describing human activities in natural language.
Annotation features include:
- Timestamped action segments
- Natural-language action descriptions
- Object references and tags
- Temporal association between recorded actions and video
Illustrative annotation examples:
- "Pick up the object from the table."
- "Place the component into the container."
- "Move the tool toward the workpiece."
These are examples of the annotation format, not claims about specific recorded activities.
The annotations support temporal action understanding, task segmentation and language-conditioned learning.
5. Operator Metadata & Consent
Associated operator information includes:
- Age
- Gender
- Dominant hand
- Years of experience
- Consent information recorded with the MCAP clip
Operator metadata provides context for analyzing human task demonstrations and variations in task execution.
Access to demographic information and consent records is subject to applicable privacy requirements and authorized usage.
6. Synchronized Multimodal Recording Structure
The dataset combines:
- Stereo RGB video
- 6-axis IMU measurements
- Sensor calibration parameters
- Video/IMU timing information
- Natural-language action segments
- Object tags
- Operator information and consent metadata
This combination supports AI systems that need to interpret visual activity, motion and language together.
Distribution
Total Available Volume: Approximately 10,000 recording hours
Data Structure: Multimodal recording episodes
Primary Container: MCAP
Stereo Video Specifications:
- Two 1920 × 1080 views
- Side-by-side 3840 × 1080 frame layout
- H.265 compression
- 30 FPS
- Global shutter
Motion Data:
- 6-axis IMU
- Approximately 500 Hz
- Time-aligned with video
Supporting Information:
- Camera intrinsics and distortion
- Stereo baseline
- Camera-to-IMU extrinsics
- Time offsets
- IMU noise profile
- Timestamped action annotations
- Object tags
- Operator metadata
Delivery Method: Custom secure cloud delivery
Dataset Selection: Flexible by available recording sessions, activity coverage and buyer requirements.
Exact file counts, storage volume, episode distribution and available recording subsets are confirmed before delivery.
Usage
Robotics & Physical AI
Training models to understand real-world human actions, movement and physical interactions from synchronized visual and inertial observations.
Vision-Language-Action Models
Supporting research into models that associate visual observations with natural-language actions and human task demonstrations.
Imitation Learning
Learning from recorded human demonstrations using stereo video and temporal action annotations.
Embodied AI
Developing AI systems capable of understanding physical activities, object interactions and spatial relationships.
Stereo Vision & Depth Estimation
Supporting stereo correspondence, disparity estimation and depth-related computer vision research using calibrated stereo imagery.
Visual-Inertial Learning
Combining stereo video with synchronized accelerometer and gyroscope measurements for motion-aware AI research.
Human Action Understanding
Training models to recognize and interpret human actions using timestamped natural-language descriptions and object tags.
Multimodal AI
Developing systems that combine vision, motion measurements and language for more comprehensive physical-world understanding.
Coverage
Data Volume: Approximately 10,000 hours
Recording Perspective: Head-mounted stereo, first-person
Data Source: Real-world human demonstrations
Activity Coverage: Human task demonstrations, with specific domains, environments and task distribution provided through the dataset inventory.
Geographic Coverage: Recording locations to be confirmed by dataset subset.
Collection Period: Varies by recording session.
Participants: Human operators with associated demographic and experience metadata.
Sensor Coverage: Stereo RGB, synchronized 6-axis IMU and calibration information.
Annotation Coverage: Timestamped natural-language action segments and object tags.
The dataset consists of real-world recordings rather than synthetic or simulated demonstrations.
License
Proprietary – Non-Exclusive Commercial Licensing
The dataset is available under non-exclusive commercial licensing, permitting TRANSFORM Solutions to license the same recordings to multiple customers.
Permitted AI training, fine-tuning, evaluation and derivative uses are governed by the agreed licensing terms and verified underlying rights.
Resale, redistribution or unauthorized sharing of original recordings, sensor data or associated metadata is prohibited without authorization.
Dataset volume, pricing, permitted uses and delivery terms are agreed before purchase.
Who Can Use It
- Physical AI Companies: Training models using multimodal human demonstrations.
- Robotics Developers: Robot learning, imitation learning and physical task understanding.
- VLA Model Developers: Training and evaluating vision-language-action systems.
- Embodied AI Research Teams: Spatial understanding and multimodal action learning.
- Computer Vision Companies: Stereo vision, depth estimation and action understanding.
- Visual-Inertial Research Teams: Sensor fusion and motion estimation research.
- AI Research Laboratories: Multimodal foundation models and human demonstration learning.
- Universities & Research Institutions: Applied robotics, computer vision and embodied intelligence research.
Data Dictionary
The fields below describe the expected recording modalities and associated metadata. Exact MCAP topic names and schemas are confirmed with the selected dataset.
| Field | Data Type | Description |
|---|---|---|
| recording_id | String | Unique recording identifier |
| stereo_video | Encoded Video | Two synchronized stereo camera views |
| video_resolution | String | 3840 × 1080 combined frame |
| camera_fps | Numeric | 30 FPS |
| video_codec | String | H.265 / HEVC |
| shutter_type | String | Global shutter |
| imu_timestamp | Timestamp | Time reference for IMU observations |
| accelerometer | Float[3] | Three-axis acceleration measurements |
| gyroscope | Float[3] | Three-axis angular velocity measurements |
| imu_frequency | Numeric | Approximately 500 Hz |
| camera_intrinsics | Matrix/Array | Intrinsic calibration parameters |
| distortion_parameters | Array | Camera distortion coefficients |
| stereo_baseline | Numeric | Approximately 60 mm |
| camera_imu_extrinsics | Matrix | Camera-to-IMU transformation |
| sensor_time_offset | Numeric | Camera/IMU synchronization offset |
| imu_noise_profile | Object | IMU noise characteristics |
| action_start_time | Timestamp | Beginning of annotated action segment |
| action_end_time | Timestamp | End of annotated action segment |
| action_description | String | Natural-language action annotation |
| object_tags | String/List | Objects referenced in an action |
| operator_age | Numeric | Operator age |
| operator_gender | String | Operator-provided gender metadata |
| dominant_hand | String | Left/right or other documented value |
| years_experience | Numeric | Operator experience |
| consent_record | Object/Reference | Associated consent information |
Additional Information
TRANSFORM Solutions offers approximately 10,000 hours of stereo egocentric video data with synchronized high-frequency IMU measurements and action annotations for commercial AI and robotics development.
Key Highlights:
- Approximately 10,000 hours of real-world stereo recordings
- Head-mounted global-shutter camera configuration
- Dual 1080p camera views at 30 FPS
- H.265 video encoding
- 6-axis IMU at approximately 500 Hz
- Video/IMU time alignment
- Stereo and camera-to-IMU calibration information
- Timestamped natural-language action annotations
- Object tags
- Operator metadata and associated consent information
- Non-exclusive commercial licensing
- Custom secure delivery
- Representative samples available to qualified buyers
Exact dataset coverage, episode counts, storage size, metadata schemas and commercial conditions are confirmed during dataset selection.
Contact TRANSFORM Solutions for technical specifications, sample recordings and licensing quotations.
TRUSTLoading...
$280,000
Download Dataset in VIDEO Format
Recommended Datasets
Loading recommendations...