Dangerous Objects Detection Using Deep Learning and First Responder Drone

Dangerous Objects Detection Using Deep Learning and First Responder Drone

Zeyad AlJundi (Naif Arab University for Security Sciences, Saudi Arabia), Saad Alsubaie (Naif Arab University for Security Sciences, Saudi Arabia), Muhammad H. Faheem (Naif Arab University for Security Sciences, Saudi Arabia), Raha Mosleh Almarashi (Naif Arab University for Security Sciences, Saudi Arabia), Emad-ul-Haq Qazi (Naif Arab University for Security Sciences, Saudi Arabia), and Jong Hyuk Kim (Naif Arab University for Security Science, Saudi Arabia)
Copyright: © 2024 |Pages: 18
DOI: 10.4018/IJDCF.367034
Article PDF Download
Open access articles are freely available for download

Abstract

Detecting dangerous objects, such as firearms or knives, is crucial for public safety or accurate situational assessment in crime scenes in law enforcement applications. Drones as first responders have been actively utilized for this purpose, showing significant benefits in law enforcement with fast and early detection of such objects. However, automated detection is still challenging, particularly with low-quality drone cameras that operate in low illumination conditions. We evaluate the performance of four popular AI deep learning models to automate the detection of dangerous objects recorded from low-quality drone cameras. The results show that the YOLOv5s model achieves the best detection performance, yielding mAP50 results of 0.964 for color and 0.949 for infrared videos, which are excellent performances considering the low-quality and low-resolution dataset. The trained network model is further implemented as an online web application where law enforcement officers can upload videos taken from drones or CCTV.
Article Preview
Top

Introduction

Drone as a first responder (DFR) has emerged as a new technological tool in security and law enforcement, in which drones are deployed to respond to an emergency call in police departments, arriving at the scenes before the ground officers while streaming live footage to the officers en route. This system enables the ground responders to prepare for what they might encounter at the scene, such as the presence of firearms. For example, the Chula Vista Police Department in the United States has adopted this technology since 2018, deploying DFR from the rooftop for over 20 thousand emergency calls, arriving at the scene within 93 seconds on average while assisting in arrests of over 28 thousand suspects (Chula Vista Police Department, 2024). Since deploying drones is typically less expensive than ground patrols or helicopters, police can allocate resources more effectively for their budgets. Although DFRs are still in their infancy, they can be utilized for crime scene investigation, disaster response, critical infrastructure protection, crowd safety, and so on.

Processing videos captured from the drone is also crucial for accurate situational assessment and decision-making in law enforcement. Most security systems have relied on camera operators' bare-eye inspection of the videos. However, this kind of task is typically very tedious and dull, as it requires processing a large volume of live or recorded footage, thus reducing the work performance. In recent years, artificial intelligence (AI) technologies have made significant progress in understanding visual data, such as object detection and scene understanding (Soori et al., 2023; Zhu et al., 2021a). Therefore, integrating this AI capability into drones can significantly enhance the performance of security systems.

Fast and robust detection of dangerous objects, like firearms and knives, is critical in the DFR systems to support the decision-makers in taking the necessary security measures. Although many datasets and AI models are available for general object detection, firearm and dangerous object datasets collected from drones are very limited. For example, gun detection datasets (Delong et al., 2021) collected gun images from internet movie databases aiming for embedded applications. Closed-circuit television (CCTV) footage is also utilized for weapon detection (Hnoohom et al., 2021), in which various deep-learning modules are evaluated. VisDrone-DET2021 releases a dataset aiming at drone object detection (Delong et al., 2021), which contains videos collected from drones over various urban areas with vehicles and pedestrians and applies deep learning models.

These public datasets are helpful but need to reflect the realistic DFR scenarios in the context of law enforcement. DFRs are typically equipped with low-cost cameras and operate from a distance for public safety. In addition, other factors, such as low-light or night-time conditions, affect the detection performance. This study collects new datasets from a realistic DFR scenario under low-light conditions to fill these gaps and evaluate AI models for detecting dangerous objects. The key contributions of this work are:

  • Collection of new dangerous objects dataset (firearms, knives) from a realistic DFR scenario.

  • Evaluation of four different AI models (“You Only Look Once” version 8 [YOLOv8], small “You Only Look Once” version 5 [YOLOv5s], faster region-based convolutional neural network [Faster R-CNN], and a visual geometry group 19 layers deep [VGG19]).

  • Development of an online application that can process uploaded videos to detect firearms and knives, aiming to support law enforcement agencies in the future in automating the evaluation of videos in security.

This is the first study on dangerous object detection from a DFR in the security field.

Complete Article List

Search this Journal:
Reset
Volume 17: 1 Issue (2025)
Volume 16: 1 Issue (2024)
Volume 15: 1 Issue (2023)
Volume 14: 3 Issues (2022)
Volume 13: 6 Issues (2021)
Volume 12: 4 Issues (2020)
Volume 11: 4 Issues (2019)
Volume 10: 4 Issues (2018)
Volume 9: 4 Issues (2017)
Volume 8: 4 Issues (2016)
Volume 7: 4 Issues (2015)
Volume 6: 4 Issues (2014)
Volume 5: 4 Issues (2013)
Volume 4: 4 Issues (2012)
Volume 3: 4 Issues (2011)
Volume 2: 4 Issues (2010)
Volume 1: 4 Issues (2009)
View Complete Journal Contents Listing