Real-time concealed object detection and recognition in passive imaging at 250  GHz

Area of Science:

• Applied Physics
• Computer Vision
• Electromagnetics

Background:

• Terahertz (THz) imaging offers unique capabilities for non-ionizing, non-contact sensing.
• Detecting concealed objects through clothing presents challenges due to fabric properties and signal attenuation.
• Existing methods require robust and high-speed algorithms for practical applications.

Purpose of the Study:

• To develop and evaluate a high-speed, robust algorithm for detecting and recognizing hidden objects concealed by various fabrics.
• To compare the performance of different deep learning approaches for concealed object detection in the THz range.
• To analyze the physical properties of fabrics relevant to THz wave penetration.

Main Methods:

• Utilized passive imagers operating in the terahertz (THz) range (1.2 mm / 250 GHz).
• Developed and trained two distinct algorithms: a multiframe single-shot detector and region-based fully convolutional networks.
• Tested algorithms using a diverse dataset of concealed objects and clothing types over 30-minute sessions.

Main Results:

• Both presented algorithms demonstrated robust detection and recognition capabilities for concealed objects.
• A comparative analysis highlighted the performance differences between the single-shot detector and fully convolutional networks.
• The study provides insights into the physical properties of fabrics affecting THz imaging.

Conclusions:

• The proposed THz imaging approach combined with deep learning offers a promising solution for concealed object detection.
• High-speed, robust algorithms are feasible for practical THz-based security and inspection applications.
• Further research can optimize these methods for enhanced accuracy and broader material characterization.