Abstract
This paper explores a multi-modal approach to security screening, specifically focusing on the challenge of identifying weapons hidden under clothing. By fusing standard color photography with infrared (IR) imaging, the researchers aim to create an automated system that provides better situational awareness than manual screening or single-sensor systems.
Introduction
Security in public spaces like airports and railway stations often relies on manual screening or single-sensor technologies, such as metal detectors. These methods have significant gaps: manual checks are slow and prone to human error, while single-sensor systems often fail to provide enough context. Visual cameras provide high-resolution detail but cannot see through fabric. Conversely, IR cameras can detect the thermal difference between a cold weapon and a warm human body, but they produce low-resolution, "noisy" images that lack the detail needed for identification.
The authors address the "Concealed Weapon Detection" (CWD) problem by proposing an algorithm that fuses high-resolution RGB visual images with thermal IR signatures using Discrete Wavelet Transform (DWT). The result is a system that can "see through" clothing to highlight the contours of a weapon while maintaining the visual context of the person carrying it.
Methodology
The proposed multi-step image processing pipeline involves the following stages:
- Preprocessing: Visual (RGB) and IR images are captured and resized to ensure pixel-to-pixel alignment.
- IR Complementation: Since weapons are typically colder than the human body, they appear as dark spots in raw IR. The authors "complement" (invert) the IR image so the weapon becomes a high-intensity (bright) feature, making it easier to extract.
- Color Space Conversion: The visual image is converted from RGB to the HSV (Hue, Saturation, Value) color space to separate color information from brightness, which is more effective for object discrimination.
- DWT Fusion: Using the Discrete Wavelet Transform, both the processed IR and visual images are decomposed into frequency bands. These bands are merged using specific fusion rules to create a composite image that retains weapon edges and visual textures.
- Post-Processing & Detection: The fused image is converted to grayscale and binarized using Otsu’s method. Area-based filtering is applied to reduce false positives by removing components that do not match the expected size of a weapon.
- Contour Mapping: Finally, a Canny edge detector identifies the boundaries of the weapon, which are then overlaid back onto the original color photo for the operator.
Key Insights and Results
The fundamental insight of this work is utilizing thermal contrast as a proxy for transparency. Because a metallic weapon has different thermal emissivity and is generally colder than the body, it creates a "thermal shadow" that the IR sensor can pick up even through fabric. Simple image addition leads to hazy results; however, by using DWT, the authors perform fusion in the transform domain, preserving sharp edges from both sources.
Qualitative results demonstrate that the system successfully isolated the shape of a handgun from a cluttered background. By mapping the detected weapon contour back onto the RGB image, the system provides a clear, high-contrast warning to security personnel without losing the identity or context of the individual being screened.
Limitations and Future Work
IR-based detection only works effectively if the clothing is relatively thin and the subject is stationary. Loose or heavy clothing, such as thick winter coats, can disperse the thermal signature, making the weapon invisible. Furthermore, the current algorithm is designed for static images. Future research will focus on adapting this pipeline to video streams, where motion blur and frame alignment present significant technical hurdles, and improving the system's ability to handle varied clothing types in high-traffic environments.
Summary updated