Enterprise AI Analysis
A Dataset for Victim Detection in Search-and-Rescue Operations Using Robot-Mounted UWB-Radar Sensors
This paper introduces a comprehensive dataset for victim detection in search-and-rescue (SAR) operations using a robot-mounted Ultra-Wideband (UWB) radar sensor. The dataset design closely emulates SAR scenarios inside buildings for victim detection and localization, capturing rich spatiotemporal data from multiple positions and angles. It addresses a critical gap in publicly available dynamic UWB radar datasets for human detection behind obstacles. Technical validity is robustly demonstrated, achieving high F1-scores of 77% (CNN) and 83% (XGBoost) for victim detection and a Mean Absolute Error (MAE) of 0.49 m for position estimation, highlighting its value for AI and ML model development in SAR.
Key Metrics & Immediate Impact
Our analysis reveals the following critical metrics and the profound impact this dataset can have on advancing autonomous SAR capabilities.
0
Total Radar Frames
0
Data Volume
0
Data Duration
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
The proposed dataset is designed to support and enhance automated robotic systems equipped with radar sensors for human detection in Search-and-Rescue (SAR) operations. It comprises a total of 2,179,281 radar frames, spanning nearly 32 hours of recording and accumulating to approximately 17 GB of data. This comprehensive collection includes measurements from 6 human subjects in various orientations and distances behind a 27 cm-thick YTONG wall, across 11 discrete robot positions. The dataset is structured to provide high-dimensional spatiotemporal data, crucial for developing robust AI/ML models that need to generalize across diverse spatial configurations.
17
GB of UWB Radar Data
Dataset Collection Workflow
Robot Position Selection (11 points)
→
Human Subject & Orientation Setup
→
Robot Scans (20cm intervals)
→
UWB Radar Data Capture (3420 frames/session)
→
Structured Dataset Output (CSV)
To validate the dataset's applicability, two distinct victim detection models—a Convolutional Neural Network (CNN) and an XGBoost model—were developed. The CNN achieved an F1-score of 77%, demonstrating its effectiveness in detecting human presence through breathing-induced radar fluctuations, especially when leveraging spatiotemporal patterns. The XGBoost model, utilizing standard deviation across range bins as features, achieved an F1-score of 83% (weighted average), showing strong performance in real-time scenarios despite not fully exploiting spatiotemporal features. Furthermore, a rule-based position estimation method, identifying the range bin with the highest standard deviation, achieved a Mean Absolute Error (MAE) of 0.49 m, providing acceptable precision for coarse localization in SAR.
| Model | Key Strengths | F1-Score (Macro) |
|---|---|---|
| CNN |
|
0.77 |
| XGBoost |
|
0.83 |
0.49
Meters MAE for Position Estimation
The dataset, while comprehensive, exhibits a class imbalance (83% presence, 17% absence). To address this and improve model generalizability, various data augmentation techniques are proposed for the absence class, including spatial shifting (±20 columns), uniform signal scaling (±20%), and adding white/pink noise. These techniques can double the effective absence data volume. Additionally, a dedicated Python library has been developed to assist users with data extraction, preprocessing (scaling, filtering), and visualization (2D heatmaps, 1D STD plots). This toolset supports advanced signal processing techniques like FFT, Wavelet analysis, and Doppler processing, streamlining research workflows.
Data Augmentation for Enhanced Model Robustness
To counteract the dataset's class imbalance and improve the generalizability of trained models across diverse environments, a series of data augmentation techniques can be applied to the 'absence' class. These include spatial shifting of radar reflections, uniform scaling of signal values, and the strategic addition of white and pink noise. These methods synthetically increase the volume of absence data without introducing bias, significantly enhancing model performance and fostering further research in artificial data generation. This approach effectively doubles the volume of absence data and leads to more robust AI models.
Impact: Doubled effective absence data, significantly boosting model generalizability and robustness.
Advanced ROI Calculator
Estimate the potential return on investment for implementing AI solutions in your enterprise based on the insights from this research.
Estimated Annual Savings
$0
Annual Hours Reclaimed
0
Implementation Roadmap
A structured approach ensures successful integration and maximum value extraction from advanced AI and data solutions.
Discovery & Strategy
Assess current infrastructure, identify key pain points, and define clear AI objectives aligned with business goals. Data readiness assessment and initial solution design.
Pilot & Development
Develop and test a proof-of-concept on a subset of data or a specific operational area. Iterate based on initial results and feedback, ensuring technical viability.
Integration & Scaling
Integrate the validated AI solution into existing enterprise systems. Scale up data pipelines and model deployment for full operational impact across relevant departments.
Performance Optimization
Continuous monitoring of AI model performance, data quality, and system efficiency. Refine algorithms and infrastructure to ensure sustained optimal outcomes.
Long-term Support
Establish ongoing maintenance, security updates, and feature enhancements. Provide training and support for end-users and adapt to evolving business needs.
Ready to Transform Your Enterprise?
Book a personalized strategy session with our AI experts to discuss how these advanced insights can be tailored to your specific business challenges and opportunities.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat