Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Gun Detection Using Combined Human Pose and Weapon Appearance

Amulya Reddy Maligireddy, Manohar Reddy Uppula, Nidhi Rastogi, Yaswanth Reddy Parla

TL;DR

This work tackles firearm detection in public spaces by fusing human pose estimation with weapon appearance recognition to provide context-aware security insights. The approach combines visual features from CNN/Transformer backbones with body-keypoint information (via Mediapipe) and employs a detector (YOLOv8-based) augmented by pose cues to improve accuracy under occlusion and clutter. A diverse dataset of 9,500–9,500+ images from IMFDB, Monash Guns, CCTV, and Roboflow, annotated in both VIA and YOLO formats, supports robust training and generalization. The final system achieves strong detection metrics (e.g., mAP50 ≈ 0.98, mAP50-95 ≈ 0.70) and includes a dynamic threat assessment that flags near-field gun placements relative to critical body parts, demonstrating practical potential for real-time surveillance. The work discusses future directions in expanded threat criteria, real-time optimization, multi-person scenarios, transformer-based alternatives, and ethical/privacy safeguards.

Abstract

The increasing frequency of firearm-related incidents has necessitated advancements in security and surveillance systems, particularly in firearm detection within public spaces. Traditional gun detection methods rely on manual inspections and continuous human monitoring of CCTV footage, which are labor-intensive and prone to high false positive and negative rates. To address these limitations, we propose a novel approach that integrates human pose estimation with weapon appearance recognition using deep learning techniques. Unlike prior studies that focus on either body pose estimation or firearm detection in isolation, our method jointly analyzes posture and weapon presence to enhance detection accuracy in real-world, dynamic environments. To train our model, we curated a diverse dataset comprising images from open-source repositories such as IMFDB and Monash Guns, supplemented with AI-generated and manually collected images from web sources. This dataset ensures robust generalization and realistic performance evaluation under various surveillance conditions. Our research aims to improve the precision and reliability of firearm detection systems, contributing to enhanced public safety and threat mitigation in high-risk areas.

Gun Detection Using Combined Human Pose and Weapon Appearance

TL;DR

This work tackles firearm detection in public spaces by fusing human pose estimation with weapon appearance recognition to provide context-aware security insights. The approach combines visual features from CNN/Transformer backbones with body-keypoint information (via Mediapipe) and employs a detector (YOLOv8-based) augmented by pose cues to improve accuracy under occlusion and clutter. A diverse dataset of 9,500–9,500+ images from IMFDB, Monash Guns, CCTV, and Roboflow, annotated in both VIA and YOLO formats, supports robust training and generalization. The final system achieves strong detection metrics (e.g., mAP50 ≈ 0.98, mAP50-95 ≈ 0.70) and includes a dynamic threat assessment that flags near-field gun placements relative to critical body parts, demonstrating practical potential for real-time surveillance. The work discusses future directions in expanded threat criteria, real-time optimization, multi-person scenarios, transformer-based alternatives, and ethical/privacy safeguards.

Abstract

The increasing frequency of firearm-related incidents has necessitated advancements in security and surveillance systems, particularly in firearm detection within public spaces. Traditional gun detection methods rely on manual inspections and continuous human monitoring of CCTV footage, which are labor-intensive and prone to high false positive and negative rates. To address these limitations, we propose a novel approach that integrates human pose estimation with weapon appearance recognition using deep learning techniques. Unlike prior studies that focus on either body pose estimation or firearm detection in isolation, our method jointly analyzes posture and weapon presence to enhance detection accuracy in real-world, dynamic environments. To train our model, we curated a diverse dataset comprising images from open-source repositories such as IMFDB and Monash Guns, supplemented with AI-generated and manually collected images from web sources. This dataset ensures robust generalization and realistic performance evaluation under various surveillance conditions. Our research aims to improve the precision and reliability of firearm detection systems, contributing to enhanced public safety and threat mitigation in high-risk areas.

Paper Structure

This paper contains 14 sections, 17 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Data
  4. Approach
  5. Architecture
  6. Methodology
  7. Baseline Object Detection Model
  8. Pose Estimation for Enhanced Detection
  9. Results
  10. Object and Human Detection
  11. Integration of Mediapipe's Pose module with YOLOv8s' object detection
  12. Conclusion
  13. Future Scope
  14. Ethical Statement

Figures (17)

  • Figure 1: Sample images from (A) Monash Guns and (B) IMFDB datasets
  • Figure 2: Samples of (A) Real- time CCTV images (B) Benign images
  • Figure 3: Samples of Images from Roboflow platfrom
  • Figure 4: Image Augmentation code snippet with parameters
  • Figure 5: False Positive correction (A) without human pose (B) with human pose information
  • ...and 12 more figures