| --- |
| language: |
| - en |
| license: other |
| task_categories: |
| - video-classification |
| - object-detection |
| tags: |
| - action-recognition |
| - human-activity-recognition |
| - industrial-ai |
| - computer-vision |
| - physical-ai |
| - har |
| - ppe-detection |
| - activity-recognition |
| - worker-safety |
| - gdpr-compliant |
| - manufacturing |
| - robotics |
| pretty_name: "ORION WWF1: Industrial AI Certified Sample Pack v2.0 (Physical AI Edition)" |
| size_categories: |
| - n<1K |
| configs: |
| - config_name: default |
| data_files: |
| - split: train |
| path: enrichments/har_annotations.csv |
| --- |
| |
| # Real industrial data, AI-ready for Physical AI |
| ## ORION WWF1 – Certified Sample Pack v2.0 (Enriched) |
|
|
| | Version | Status | Sector | Pipeline | |
| | :--- | :--- | :--- | :--- | |
| | **v2.0-Enriched** | 🟢 **Level 3 Certified** | Industrial-Manufacturing | Orion Unified V5.2 | |
|
|
| --- |
|
|
| ## 🌟 The Evolution: Beyond Anonymization |
| The **ORION WWF1 v2.0 Enriched** pack represents the professional evolution of our baseline industrial dataset. While previous versions focused on privacy-first anonymization, v2.0 transforms raw video into actionable, multi-layered intelligence for **Physical AI**. |
|
|
| This pack includes 10 representative clips from a high-intensity wood-processing facility, now featuring **Certified Human-in-the-Loop (HITL) annotations** across four distinct enrichment layers: Safety Compliance (PPE), Activity Recognition (HAR), Industrial Environment (Physical AI), and Machine Learning Detection (COCO). |
|
|
| ## 📸 Visual Preview (Certified Annotations) |
|
|
| | Clip SRC_021 | Clip SRC_029 | Clip SRC_039 | |
| | :---: | :---: | :---: | |
| |  |  |  | |
| | **PPE & HAR Audit** | **Physical AI Context** | **Operational Logic** | |
| |
| --- |
| |
| ## 1. Dataset Summary & Mission |
| **ORION WWF1** is a specialized dataset designed for the next generation of **Physical AI** and **Industrial Safety** models. It captures authentic, complex worker-machine interactions in a Wooden Window Factory (WWF1). |
| |
| * **Mission**: To provide high-fidelity, privacy-compliant data for training models that understand complex industrial workflows, hazardous environments, and safety protocols. |
| * **Enrichment**: Every frame is enriched with spatial and temporal metadata, audited by human experts to ensure "Zero False Negative" reliability and absolute privacy. |
| * **Certification**: Every asset is **Level 3 Certified**, meaning it has passed a dual-pass AI anonymization process followed by a 100% manual human audit. |
| |
| ## 2. Enrichment Layers (The "Multi-Layer Stack") |
| |
| ### 🟢 Layer 1: HAR (Human Activity Recognition) |
| Full temporal sequences of worker actions, essential for cycle-time analysis and workflow optimization. |
| * **Format**: `har_annotations.csv` |
| * **Labels**: `cutting_wood`, `carrying_object`, `assembling`, `operating_machine`, `standing_idle`, `walking`, `inspecting`, `cleaning`. |
| * **Accuracy**: Audited at 1.0s temporal resolution with precise start/end timestamps. |
|
|
| ### 🔵 Layer 2: PPE (Safety Compliance) |
| Multi-zone equipment verification per worker, compliant with international HSE standards. |
| * **Format**: `ppe_summary.csv` + `compliance_report.html` |
| * **Zones Monitoring**: Head (Hardhat), Eye (Glasses), Hearing, Respiratory, Hand (Gloves), Body (Vest), Legs, Foot (Safety shoes). |
| * **Standard**: 8-category compliance audit (YES / NO / HIDDEN). |
|
|
| ### 🟣 Layer 3: PHYSICAL AI (Environmental Context) |
| Deep semantic understanding of the industrial workspace for robotics and digital twin training. |
| * **Format**: `physical_ai_environment.json` |
| * **Data**: Machine types, floor conditions (concrete), navigation hazards, camera POV parameters (height/angle). |
| * **Geography**: Documented industrial hub in Romania. |
|
|
| ### 🟡 Layer 4: ML TRAINING (COCO Annotations) |
| High-quality keyframes for object detection and computer vision training. |
| * **Format**: `coco_annotations.json` |
| * **Details**: 310 certified keyframes with high-density bounding boxes. |
| * **Privacy**: All frames are extracted *after* anonymization to guarantee 100% privacy compliance. |
|
|
| ## 3. Dataset Structure (v2.0 Premium Layout) |
| ```text |
| ORION_WWF1_v2.0_ENRICHED/ |
| ├── ORION_WWF1_SAMPLES_v2.zip # Single-click download for the full pack |
| ├── samples/ # Final MP4 Deliverables (1080p, Dual-Pass Anonymized) |
| ├── enrichments/ # 🆕 All Intelligence Layers (CSV, JSON, HTML) |
| │ ├── har_annotations.csv # Activity sequences |
| │ ├── ppe_summary.csv # Worker safety records |
| │ ├── physical_ai_environment.json # Scene context |
| │ └── coco_annotations.json # ML-ready Bounding Boxes |
| ├── previews/ # High-resolution visual thumbnails |
| ├── integrity/ # SHA-256 Master Manifest & Checksums |
| ├── metadata/ # Unified Field Specs & Pipeline Schema |
| ├── DATASHEET.md # Technical Deep-Dive & Ethics |
| ├── METHODOLOGY.md # Dual-Pass Anonymization & HITL Process |
| └── README.md # This documentation |
| ``` |
|
|
| ## 4. Collection & Methodology |
| Data was collected using calibrated industrial cameras during active production cycles. |
| * **Environment**: Real woodworking factory (high ambient noise, industrial dust, variable lighting). |
| * **Anonymization**: Our proprietary **Dual-Pass Pipeline (V4.2 Industrial)** applies irreversible Gaussian Anonymization to all human subjects, ensuring zero-leak privacy. |
| * **Verification**: Secondary scan for residual leaks using Vertical/Horizontal Flip TTA. |
|
|
| ## 5. Annotation & HITL Process |
| Unlike purely automated datasets, the ORION v2.0 utility follows our proprietary **Adjudicator Workflow**: |
| 1. **AI Predetection**: Specialized models propose initial BBoxes and Activity segments. |
| 2. **Human-in-the-loop (HITL)**: 100% manual review and correction of every single frame and sequence by industrial analysts. |
| 3. **QA Certification**: Final signature verifying zero-leak privacy and 100% annotation integrity. |
|
|
| ## 6. Certified Feature Metadata |
| | Feature | Status | Specification | |
| | :--- | :--- | :--- | |
| | **Clip Count** | 10 Clips | HD 1080p, 59.94 fps | |
| | **Anonymization** | 🟢 **Dual-Pass** | Faces & Heads (Irreversible Blur) | |
| | **BBox Annotations** | ✅ Included | Professional COCO JSON format | |
| | **PPE Detection** | ✅ Certified | 8-Zone Compliance Audit | |
| | **HAR Actions** | ✅ Certified | Temporal localization (HITL verified) | |
| | **Integrity** | ✅ Verified | Full SHA-256 Recursive Manifest | |
|
|
| ## 7. Use Cases |
| * **Safety Analysis (HSE)**: Building models for automated safety non-compliance alerts. |
| * **Production Efficiency**: Cycle time analysis and bottleneck identification. |
| * **Robotics & Digital Twins**: Training autonomous agents to navigate and interact with industrial assets. |
|
|
| ## 8. Integrity & Licensing |
| * **Integrity**: Every file is signed via SHA-256 (see `integrity/manifest.sha256`). |
| * **License**: Distributed under the **ORION Sample Data License (v2.0 Enriched)**. Commercial R&D use permitted. |
|
|
| 🚀 **About ORION** |
| **ORION – Industrial AI Data Lab** is the bridge between industrial privacy and AI performance. We deliver "Small Data with Big Impact" for the manufacturing sector. |
|
|
| Explore more: [orion-the-lab.com](https://orion-the-lab.com) | [LinkedIn](https://www.linkedin.com/company/orion-the-lab) |
|
|
