Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

UAV-Enhanced Combination to Application: Comprehensive Analysis and Benchmarking of a Human Detection Dataset for Disaster Scenarios

Ragib Amin Nihal, Benjamin Yen, Katsutoshi Itoyama, Kazuhiro Nakadai

TL;DR

This work addresses the lack of disaster-specific human-detection data for UAV-based SAR by introducing the C2A dataset, a synthetic resource created by overlaying diverse human poses onto disaster-scene backgrounds. The authors develop a full dataset creation pipeline that combines AIDER disaster backgrounds with LSP/MPII-MPHB poses, producing 10,215 images with over 360,000 instances and rich pose/context annotations. Benchmarking across multiple state-of-the-art detectors shows substantial gains when models are fine-tuned on C2A, and further improvements when C2A is combined with general human datasets, highlighting the value of domain-specific data for robust SAR performance. The work also analyzes object-size effects, model generalization, and limitations of synthetic data, suggesting future directions including video-based sequences and real-world validation to bridge the synthetic-real gap for practical, AI-assisted disaster response.

Abstract

Unmanned aerial vehicles (UAVs) have revolutionized search and rescue (SAR) operations, but the lack of specialized human detection datasets for training machine learning models poses a significant challenge.To address this gap, this paper introduces the Combination to Application (C2A) dataset, synthesized by overlaying human poses onto UAV-captured disaster scenes. Through extensive experimentation with state-of-the-art detection models, we demonstrate that models fine-tuned on the C2A dataset exhibit substantial performance improvements compared to those pre-trained on generic aerial datasets. Furthermore, we highlight the importance of combining the C2A dataset with general human datasets to achieve optimal performance and generalization across various scenarios. This points out the crucial need for a tailored dataset to enhance the effectiveness of SAR operations. Our contributions also include developing dataset creation pipeline and integrating diverse human poses and disaster scenes information to assess the severity of disaster scenarios. Our findings advocate for future developments, to ensure that SAR operations benefit from the most realistic and effective AI-assisted interventions possible.

UAV-Enhanced Combination to Application: Comprehensive Analysis and Benchmarking of a Human Detection Dataset for Disaster Scenarios

TL;DR

This work addresses the lack of disaster-specific human-detection data for UAV-based SAR by introducing the C2A dataset, a synthetic resource created by overlaying diverse human poses onto disaster-scene backgrounds. The authors develop a full dataset creation pipeline that combines AIDER disaster backgrounds with LSP/MPII-MPHB poses, producing 10,215 images with over 360,000 instances and rich pose/context annotations. Benchmarking across multiple state-of-the-art detectors shows substantial gains when models are fine-tuned on C2A, and further improvements when C2A is combined with general human datasets, highlighting the value of domain-specific data for robust SAR performance. The work also analyzes object-size effects, model generalization, and limitations of synthetic data, suggesting future directions including video-based sequences and real-world validation to bridge the synthetic-real gap for practical, AI-assisted disaster response.

Abstract

Unmanned aerial vehicles (UAVs) have revolutionized search and rescue (SAR) operations, but the lack of specialized human detection datasets for training machine learning models poses a significant challenge.To address this gap, this paper introduces the Combination to Application (C2A) dataset, synthesized by overlaying human poses onto UAV-captured disaster scenes. Through extensive experimentation with state-of-the-art detection models, we demonstrate that models fine-tuned on the C2A dataset exhibit substantial performance improvements compared to those pre-trained on generic aerial datasets. Furthermore, we highlight the importance of combining the C2A dataset with general human datasets to achieve optimal performance and generalization across various scenarios. This points out the crucial need for a tailored dataset to enhance the effectiveness of SAR operations. Our contributions also include developing dataset creation pipeline and integrating diverse human poses and disaster scenes information to assess the severity of disaster scenarios. Our findings advocate for future developments, to ensure that SAR operations benefit from the most realistic and effective AI-assisted interventions possible.
Paper Structure (24 sections, 3 figures, 5 tables, 1 algorithm)

This paper contains 24 sections, 3 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Dataset Creation Pipeline
  4. Data Sources and Composition
  5. Pipeline Steps
  6. 1. Background Removal and Image Preparation:
  7. 2. Image Cropping and Cleaning:
  8. 3. Overlay Process:
  9. Properties of C2A Dataset
  10. Number of Images and Image Size
  11. Objects Size
  12. Aspect Ratio of Objects
  13. Object Density of Images
  14. Human Pose and Disaster Scene Information
  15. Evaluation
  16. ...and 9 more sections

Figures (3)

  • Figure 1: This collection of images presents a selection from our proposed "Combination to Application" (C2A) dataset, a specialized compilation designed to refine machine learning algorithms for SAR operations in diverse disaster scenarios. Within the bounding boxes, human figures are superimposed onto various disaster scenes, demonstrating the intricate process of overlaying accurately segmented human poses onto different disaster backdrops such as rubble, traffic incidents, flood, and fire. This synthetic approach is crucial for creating more challenging training conditions that AI models may encounter in actual SAR missions. Furthermore, the dataset is enriched with detailed pose information—such as bent, kneeling, lying, sitting, and upright—providing comprehensive data for AI to learn and recognize human forms even when partially occluded by environmental obstacles.
  • Figure 2: (a) Aspect Ratio of C2A Dataset (b) Object Density
  • Figure 3: Comparative Analysis of Object Detection (a) The frequency distribution of ground truth object sizes (blue) showcases a clear decline in detection rates for smaller objects (red), highlighting the challenges current detection algorithms face with objects less than 20 pixels in size. (b) The detection confidence scores across varying object sizes, with mean confidence indicated by red points, emphasize the higher reliability of detecting larger objects. These visualizations underscore the need for refining detection algorithms to better recognize small objects, which are critical for comprehensive disaster scene analysis.