Source-linked AI summary
Intelligent Metasurface Imager and Recognizer
Lianlin Li, Ya Shuang, Qian Ma, Haoyang Li, Hanting Zhao, Menglin Wei1, Che Liu, Chenglong Hao, Cheng-Wei Qiu, Tie Jun Cui
TL;DR
Remote monitoring requires systems that can operate without deliberate cooperation or identification tags while handling successive tasks in real time. The paper proposes an ANNs-driven intelligent metasurface imager-recognizer and demonstrates robust monitoring of multiple non-cooperative people, including under passive excitation by stray Wi-Fi signals.
Problem
Remote monitoring must probe where people are and what they express through body signals, but conventional systems typically require cooperation or devices and are designed for only one specific task.
Method
An ANNs-driven intelligent metasurface uses a large-aperture programmable surface for adaptive electromagnetic-wave manipulation and smart data acquisition.
Results
The system robustly monitored notable movements, subtle gestures, and physiological states of multiple non-cooperative people in real-world settings, including under passive excitation by stray Wi-Fi signals.
Takeaways & Limitations
The demonstrated strategy supports real-world smart-city, smart-home, interactive-interface, health-monitoring, and safety-screening applications without visual privacy issues.
Takeaways & Limitations
Conventional systems are typically limited to one specific task and hardly perform successive missions adaptively.
Abstract
from arXiv · showhide
It is ever-increasingly demanded to remotely monitor people in daily life using radio-frequency probing signals. However, conventional systems can hardly be deployed in real-world settings since they typically require objects to either deliberately cooperate or carry a wireless active device or identification tag. To accomplish the complicated successive tasks using a single device in real time, we propose a smart metasurface imager and recognizer simultaneously, empowered by a network of artificial neural networks (ANNs) for adaptively controlling data flow. Here, three ANNs are employed in an integrated hierarchy: transforming measured microwave data into images of whole human body; classifying the specifically designated spots (hand and chest) within the whole image; and recognizing human hand signs instantly at Wi-Fi frequency of 2.4 GHz. Instantaneous in-situ imaging of full scene and adaptive recognition of hand signs and vital signs of multiple non-cooperative people have been experimentally demonstrated. We also show that the proposed intelligent metasurface system work well even when it is passively excited by stray Wi-Fi signals that ubiquitously exist in our daily lives. The reported strategy could open a new avenue for future smart cities, smart homes, human-device interactive interfaces, healthy monitoring, and safety screening free of visual privacy issues.
Introduction
The paper targets real-time, privacy-preserving monitoring of non-cooperative people, addressing systems that are task-specific, computationally difficult, and hardware-intensive. It proposes an ANN-driven programmable metasurface for full-scene imaging, adaptive local sensing, and recognition using microwave signals.
- Motivation: Existing RF systems can locate people, recognize actions and poses, and monitor breathing without active devices or tags, but typically support only one specific task.Successive adaptive missions such as searching for people, focusing on body regions, and recognizing signs remain difficult.
- Motivation: Monitoring hand signs and vital signs is difficult because people may need to cooperate and weak signals can be confused with disturbances.The challenge includes local gesture language, respiration, and heartbeat in real-world settings.
- Motivation: Conventional approaches also incur complicated designs and high hardware costs because they use many transmitters and receivers.The paper motivates an inexpensive device that can obtain high-resolution whole-body images and adaptively recognize body and vital signs.
- Approach: The proposed system combines a large-aperture programmable metasurface with three CNNs for real-time end-to-end mappings from microwave data to images and recognition results.The metasurface adaptively manipulates electromagnetic waves and supports smart data acquisition and processing.
- Approach: The metasurface is designed to image multiple people, focus electromagnetic fields on selected local body spots, and monitor body and vital signs by scanning regions of interest.The system supports both ambient Wi-Fi signals and local adaptive sensing.
- Demonstration: At 2.4 GHz, experiments demonstrate high-resolution full-scene imaging, recognition of body language and respiration, and operation under passive stray Wi-Fi excitation.The reported system is intended for real-time, inexpensive monitoring without visual privacy intrusion.
Results
The intelligent metasurface combines adaptive electromagnetic control with neural-network processing to image people and recognize hand signs and respiration in real time. Experiments demonstrate high-resolution imaging, through-wall detection, focused sensing, and accurate recognition for multiple people, including under stray Wi-Fi illumination.
- Adaptive sensing: 20 dB enhancement in echo SNR focuses Wi-Fi sensing on desired body parts and supports subsequent hand-sign and vital-sign recognition.The metasurface adaptively focuses electromagnetic waves while suppressing unwanted disturbance and clutter.
- Whole-body imaging: The system detects notable movements of test persons behind a 5 cm-thick wooden wall and produces images of multiple people behind obstacles.The experiments validate the system’s see-through-the-wall capability in a laboratory environment.
- Whole-body imaging: 53 coding patterns produce high-quality images, while 63 patterns keep total data acquisition below 0.7 ms.The metasurface uses frequency points from 2.4 to 2.5 GHz, with coding-pattern switching taking around 10 μs.
- Recognition performance: Hand-sign recognition exceeds 95% accuracy, while respiration identification reaches 95% and beyond for two test persons.Normal breathing and breath holding are clearly distinguished, and hand-sign accuracy is largely unaffected by the number of test persons after hand localization.
Conclusions
The paper presents an ANNs-driven intelligent metasurface for remotely monitoring movements, gestures, and physiological states of multiple non-cooperative people. Experiments show operation with commodity Wi-Fi signals, while the authors identify higher resolution and accuracy as needed for broader recognition tasks.
- Conclusions: The system robustly monitors notable movements, subtle body-gesture languages, and physiological states of multiple non-cooperative people in real-world settings.The approach combines a large-aperture programmable metasurface for adaptive wave manipulation with three ANNs for real-time data-flow processing.
- Conclusions: Passive excitation by commodity Wi-Fi signals still supports monitoring movements of non-cooperative persons in the real world.The authors experimentally demonstrated that the intelligent metasurface works well under passive Wi-Fi illumination.
- Conclusions: The strategy is presented as relevant to assisting handicapped people and supporting body-language-based device control.The paper specifically mentions remotely sending commands to devices using body languages.
- Conclusions: Lip reading and mood recognition are framed as possible extensions requiring higher resolution and accuracy, potentially through higher frequencies.The paper states that the intelligent-metasurface concept can in principle extend across the electromagnetic spectrum.
- Conclusions: The authors identify smart homes, human-device interfaces, healthy monitoring, and safety screening as future application areas.These applications are described as avenues opened by the reported strategy.
Methods
The method integrates a programmable metasurface with three convolutional neural networks to acquire, reconstruct, localize, and classify microwave information. Its active and passive configurations use different illumination and receiving arrangements, while FPGA-controlled coding patterns adapt electromagnetic wavefields to target body regions.
- Metasurface design: The programmable metasurface contains 32 × 24 electronically controlled meta-atoms operating around 2.4 GHz.Each meta-atom uses a PIN diode for electronic control.
- Metasurface design: CST Microwave Studio models PIN-diode states, x-polarized incidence, reflected waves, and periodic boundaries for an infinite-array approximation.The design procedure investigates the meta-atom’s reflection response under different diode states.
- System configuration: Active operation uses transmitting and receiving antennas with a vector network analyzer, whereas passive operation uses receiving antennas and an oscilloscope.One passive receiver serves as a reference for calibrating undesirable system-error effects.
- Adaptive acquisition: FPGA-controlled coding sequences make the metasurface both an electronically controllable random mask and a focusing element for desired body spots.The focusing patterns suppress irrelevant inferences and clutter while transferring specimen information to the receivers.
- Neural-network processing: IM-CNN-1 converts microwave raw data into whole-body images, Faster R-CNN identifies hands and chests, and IM-CNN-2 infers hand signs from microwave data.The three networks form the core real-time data-processing hierarchy.
- Neural-network processing: IM-CNN-1 and IM-CNN-2 operate directly on microwave raw data and are trained using ADAM with mini-batches of 32 and 101 epochs.The learning rates are 10^-4 and 10^-5 for the first two layers and the last layer, respectively.