Our firearms detection technology utilizes AI to enhance threat detection by identifying firearms and localizing firearm carriers. We explore the forefront of AI-powered security, where innovation meets public safety through proactive firearm detection with advanced algorithms.
a. Detection and Localization of Firearm Carrier Persons
We develop an innovative approach for detecting and accurately localizing
individuals carrying firearms in images or videos, specifically addressing
challenges posed by clutter and diverse firearm shapes in complex environments.
Our proposed method leverages human–firearm interaction information and
incorporates an attention mechanism to discern relevant areas within the images
or videos.
A saliency-driven locality-preserving constraint is introduced to facilitate learning
essential features while preserving foreground information. The utilization of
attention mechanisms and saliency-driven locality preservation marks significant
advancements in firearm localization and identification, offering enhanced
capabilities for security and surveillance applications.
A dataset containing thousands of annotated firearm carrier images is also
prepared.
Related Publications
b Visual Firearms Detection and Localization
We develop an innovative algorithm for automatic firearm detection, specifically
addressing challenges posed by variations in shape, size, and appearance of
firearms. In contrast to existing object detectors, we use axis-aligned bounding
boxes for training, and we predict oriented bounding boxes during the detection
phase. We have also developed a firearm training dataset containing around 6000
annotated firearm images.
Related Publications
c. Improving security surveillance by hidden cameras
In this study our focus will be to develop the text highlights challenges in security
surveillance, particularly the vulnerability of traditional cameras due to their
visible locations, which can be exploited by criminals. To address this issue, the
paper proposes an innovative solution involving hidden cameras strategically
positioned to capture video through narrow regions, such as window curtain slits.
The presented framework is effective and robust, automatically extracting slit
regions and seamlessly merging them across frames to construct a panoramic
view. The system successfully tackles challenges such as sudden illumination
variations, demonstrating its effectiveness through experiments on various real
video sequences. Importantly, it outperforms existing commercial software in the
perceptual quality of the resulting panoramic images, showcasing the potential of
this solution for enhancing security surveillance in discreet and effective ways.
Related Publications