Title: The application of AI techniques for firearm detection in digital forensic investigation

Authors: Suraj Harsha Kamtam; Harjinder Singh Lallie; Muhammad Ajmal Azad

Addresses: CFTC, Coventry University, Priory St., Coventry, CV1 5FB, UK ' WMG, University of Warwick, Coventry, CV4 7AL, UK ' School of Computing and Digital Technology, Birmingham City University, Belmont Row, Birmingham B4 7RQ, UK

Abstract: The early detection of potentially violent situations involving firearms is a useful aid to law enforcement. AI and automation can complement humans in weapon detection as they excel in repetitive tasks and make clear judgments of ambiguous situations. AI technology can be used in digital forensic investigations to detect objects such as firearms and predict features such as age, and gender. This paper demonstrates the application of a model called you only look once (YOLOv3), a deep neural network, which was used to build a custom firearm detection model. The proposed model can automate the repetitive, tedious and error-prone task of searching through a large number of images for the presence of firearms, thus reducing human effort and stress. Five models have been trained in this paper on different scenarios to understand the performance of YOLOv3 which include one firearm, multiple firearms (pistol and rifle), greyscale images, factual scene and L-shape false positives. Our model achieved a maximum mean average precision of 97.68% and a minimum of 59.41%. The models developed in this work outperform existing models which do not scale well and cannot detect changes in image, noise, shape and background.

Keywords: digital forensics? video forensics? you only look once? YOLO? YOLOv3? firearm detection? transfer learning? computer vision? convolutional neural network? CNN? IMFDB? crime prediction? crime analysis? deep learning? crime scenes? violence detection.

DOI: 10.1504/IJESDF.2024.138343

International Journal of Electronic Security and Digital Forensics, 2024 Vol.16 No.3, pp.372 - 396

Received: 01 Jul 2022
Accepted: 10 Jan 2023
Published online: 01 May 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article