Browse Prior Art Database

Anomalous Object detection in mm-wave imagery

IP.com Disclosure Number: IPCOM000200693D
Publication Date: 2010-Oct-25
Document File: 5 page(s) / 265K

Publishing Venue

The IP.com Prior Art Database

Abstract

An urgent and increasing demand exists for new technologies and capabilities for the detection of concealed weapons and threats. With the advent of passive millimeter wave (MMW) imaging devices such as Brijot’s GEN 2, it is possible to make visible anomalous objects such as weapons or explosives that would ordinarily be concealed underneath clothing or other visual obstructions. The invention is a new system for rapidly processing MMW imagery to robustly detect a wide range of anomalous objects and utilizes recent advancements in several related fields such as codebook generation using vector quantization, object recognition using classifier cascades with swarm-based search, and mean-shift tracking to detect anomalous objects. The codebook generation helps to reduce the inherent noise in the data by quantizing the information so the dominant modes are preserved. The object recognition uses advances in combining evolutionary search mechanisms with statistical classifier cascades to rapidly detect and localize the objects of interest. The mean-shift based tracking helps to reliably track objects using intensity histograms and computationally inexpensive similarity functions. The innovative combination of these techniques helps in automatically detecting anomalous objects, preserving privacy, and obviating the need for constraining the subject so he can be scanned.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 38% of the total text.

Page 01 of 5

An urgent and increasing demand exists for new technologies and capabilities for the detection of concealed weapons and threats. With the advent of passive millimeter wave (MMW) imaging devices such as Brijot's GEN 2, it is possible to make visible anomalous objects such as weapons or explosives that would ordinarily be concealed underneath clothing or other visual obstructions. Such systems have been shown to have tremendous potential in the areas of portal checkpoint security and stand off threat detection. A need exists for automated methods to detect anomalous objects both to improve the efficiency of such systems and to obviate privacy concerns. Automated detection of weapons and explosives hidden under people's clothing has many security applications. Its advantages include reducing risks to security personnel, reducing manpower requirements, and causing less social friction since it reduces the need for hands-on searching of individuals. DARPA programs such as SPEYES are very interested in solving the hidden weapon problem. However, even the most sophisticated MMW sensor technology yields low resolution images with noise that clutters the image and can easily confuse conventional object detection algorithms. Along with rapid strides in passive MMW sensor technologies, it is imperative to look at image analysis techniques that have the potential to improve object detection by increasing the amount of contextual information elicited from the data. Our invention is a new technique for rapidly processing MMW imagery to robustly detect a wide range of anomalous objects.

Our invention describes a robust approach to reliably detect anomalous objects in MMW imagery. Our approach uses recent advancements in several related fields such as codebook generation using vector quantization, object recognition using classifier cascades with swarm-based search, and mean-shift tracking to detect anomalous objects in unconstrained environments while preserving privacy. The codebook generation helps to reduce the inherent noise in the data by quantizing the information so the dominant modes are preserved. The object recognition uses advances in combining evolutionary search mechanisms with statistical classifier cascades to rapidly detect the objects of interest. The mean-shift based tracking helps to reliably track objects using intensity histograms and computationally inexpensive similarity functions. A block diagram of our proposed approach is shown in Figure 1.

(This page contains 00 pictures or other non-text object)

Figure 1. A block diagram of our approach for anomaly detection in mm-wave imagery.

The input mm-wave imagery is first passed to a texture-based pre-processing module. Several texture measures based on the second order pixel statistics are computed and


Page 02 of 5

used to determine if the frame in question contains a humans subject or not. Examples of the texture measures extracted for an example sequence collected using the...