Daniel Chen,Anton Schlegel,Jeffrey A. Nanzer
Abstract:We demonstrate an imageless method of concealed contraband detection using a real-time 75 GHz rotationally dynamic antenna array. The array measures information in the two-dimensional Fourier domain and captures a set of samples that is sufficient for detecting concealed objects yet insufficient for generating full image, thereby preserving the privacy of screened subjects. The small set of Fourier samples contains sharp spatial frequency features in the Fourier domain which correspond to sharp edges of man-made objects such as handguns. We evaluate a set of classification methods: threshold-based, K-nearest neighbor, and support vector machine using radial basis function; all operating on arithmetic features directly extracted from the sampled Fourier-domain responses measured by a dynamically rotating millimeter-wave active interferometer. Noise transmitters are used to produce thermal-like radiation from scenes, enabling direct Fourier-domain sampling, while the rotational dynamics circularly sample the two-dimensional Fourier domain, capturing the sharp-edge induced responses. We experimentally demonstrate the detection of concealed metallic gun-shape object beneath clothing on a real person in a laboratory environment and achieved an accuracy and F1-score both at 0.986. The presented technique not only prevents image formation due to efficient Fourier-domain space sub-sampling but also requires only 211 ms from measurement to decision.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: on the premise of ensuring personal privacy, develop a method for detecting hidden contraband (such as metal gun - like objects) without generating images. Specifically, this method uses a 75 - GHz rotating dynamic antenna array to sample in the spatial Fourier domain, and realizes the detection of hidden objects through a small number of Fourier samples, which are not sufficient to reconstruct a complete image, thus protecting the privacy of the detected person.
### Main problems and solutions
1. **Privacy protection**:
- Traditional image - based contraband detection methods will expose the sensitive information of the detected person.
- The method proposed in the paper only collects enough Fourier - domain samples for detection without generating a complete image, thus avoiding privacy leakage.
2. **Efficiency improvement**:
- Traditional methods need to generate a complete image and then apply image - processing techniques for detection and classification, which is not only time - consuming, and usually the contraband only accounts for a small part of the image.
- The new method reduces the amount of data and processing time and improves the detection efficiency by directly sampling and classifying in the Fourier domain.
3. **Detection accuracy**:
- The paper uses multiple classification algorithms (such as the threshold method, K - nearest neighbor and support vector machine), and directly extracts arithmetic features from the collected Fourier - domain responses for classification.
- The experimental results show that this method can detect metal gun - like objects hidden under clothing with high precision (both the accuracy rate and F1 - score are 0.986) in the laboratory environment.
### Method overview
- **System design**: Use a 75 - GHz rotating dynamic antenna array, combined with active incoherent millimeter - wave (AIM) technology, irradiate the scene through a noise emitter to make the radiation signal uncorrelated in space - time.
- **Fourier - domain sampling**: The dynamic antenna array collects spatial Fourier - domain samples during the rotation process, and these samples contain sharp spatial - frequency features caused by object edges.
- **Classification algorithm**: Use algorithms such as the threshold method, K - nearest neighbor and support vector machine to classify the extracted features to achieve real - time detection.
### Formula representation
- Spatial Fourier transform relationship:
\[
V(u,v)=\int_{-\infty}^{+\infty}I(l,m)e^{j2\pi(ul + vm)}\,dl\,dm
\]
where \(V(u,v)\) is the scene visibility, \(I(l,m)\) is the scene intensity, and \(l\) and \(m\) represent the direction cosines of the azimuth and elevation angles respectively.
- Inverse transform formula:
\[
I_r(l,m)=\int_{-\infty}^{+\infty}V_s(u,v)e^{-j2\pi(ul + vm)}\,du\,dv
\]
where \(V_s(u,v)\) is the sampled visibility.
Through the above methods, this research not only realizes efficient contraband detection but also ensures that the privacy of the detected person is not violated.