Human activity and motion detection are vital for applications like remote healthcare, intrusion detection, and independent living, a key component of the UK's 2030 national agenda. Existing systems using ambient sensors, cameras, and wearables raise privacy and comfort concerns. Contactless monitoring using microwave sensing, exploiting channel state information (CSI) or Doppler signatures, offers a solution. However, current microwave sensing systems face limitations: weak reflection signals restrict range to a few meters; environmental interference degrades performance in Non-LOS scenarios where transmitter (Tx) and receiver (Rx) lack a direct link; and most schemes require controlled, pseudo-dynamic movement settings. This work addresses these limitations by introducing Intelligent Wireless Walls (IWW). IWW leverage reconfigurable intelligent surfaces (RIS) for beamforming and machine learning for activity recognition. RIS dynamically control electromagnetic (EM) wave steering, enhancing sensing range and mitigating interference. The IWW concept combines the advantages of RIS (low complexity, scalability, and low power consumption) with AI for high-precision, contactless activity monitoring in complex environments.

Several existing contactless human activity monitoring systems utilize microwave sensing based on WiFi, 5G, or radar. However, these systems are limited by factors such as short detection range due to weak reflected signals in Non-LOS environments and significant interference from the surrounding environment. Recent research has explored the use of RIS in object and gesture recognition, microwave imaging, and smart metasurface imaging. Non-LOS imaging techniques using acoustics, long-wave infrared (IR), and cameras have been developed, but they may have limitations such as requiring multiple measurements, raising privacy concerns, or being dependent on lighting conditions. This study builds upon these existing approaches by integrating RIS into a microwave sensing system to overcome the challenges of Non-LOS environments, improving the accuracy of activity recognition.

The proposed IWW system uses an RIS to steer microwave signals towards the target, improving the signal-to-noise ratio and reducing interference. The system uses a machine learning approach to classify different human activities. Two scenarios were considered: a corridor junction scenario with Tx and Rx in separate corridors, and a multi-floor scenario with Tx and Rx on different floors. Three activities (sitting, standing, walking) were performed by two subjects (male and female) in each scenario. Three machine learning algorithms (Random Forest (RF), Extra Trees (ET), and Multilayer Perceptrons (MLP)) were evaluated using both test-train split and repeated stratified k-fold cross-validation techniques. The performance of the IWW system was compared to a conventional microwave sensing system without RIS. The RIS testbed used in the study has a high resolution beam-steering capability in the azimuthal plane, providing near-3-bit phase resolution. The unit cell design consists of copper patches, PIN diodes, and a capacitor on a grounded dielectric substrate. The phase response of each unit cell is controlled by varying the biasing states of the PIN diodes.

The results demonstrate that the IWW system significantly improves the accuracy of activity recognition compared to the conventional system without RIS. In the corridor junction scenario, the accuracy gain ranged from approximately 20% to 25%, reaching 100% accuracy for some algorithms and subjects with the RIS enabled. The multi-floor scenario showed even greater improvement, with accuracy gains exceeding 28% in some cases. The confusion matrices (Figures 1 and 2) illustrate the improved classification accuracy with RIS enabled, showing a dramatic reduction in misclassifications for all activities. Figure 3 shows that the use of RIS consistently increased the detection accuracy across all algorithms and scenarios. The maximum accuracy improvement is observed in the multi-floor scenario using the ET algorithm. This improvement is attributed to the beamforming capabilities of the RIS, which enhances the reflection signal from the subject, resulting in improved activity detection. The detailed accuracy values are presented in Tables 1-4 for both scenarios and evaluation methods.

The findings demonstrate the effectiveness of IWW in overcoming the limitations of conventional microwave sensing for contactless activity monitoring. The use of RIS significantly enhances the signal quality in Non-LOS environments, leading to improved accuracy in activity classification. The results support the potential of IWW as a robust and privacy-preserving solution for in-home monitoring, particularly beneficial for assisting independent living among the elderly and disabled population. The high accuracy achieved in both complex scenarios highlights the versatility and adaptability of the proposed system. The improvement in accuracy is directly attributed to the RIS's capability to focus the microwave signal towards the subject, thereby increasing the signal strength and reducing interference. This work provides a significant advancement in contactless activity monitoring technology.

This paper presented the concept of Intelligent Wireless Walls (IWW) for contactless in-home activity monitoring. By integrating reconfigurable intelligent surfaces (RIS) and machine learning, IWW successfully addressed the limitations of conventional microwave sensing in complex Non-LOS environments. Experimental results in corridor junction and multi-floor scenarios showed significant accuracy gains, demonstrating the potential of IWW for robust and privacy-preserving activity monitoring. Future work could explore the integration of more sophisticated machine learning models, larger RIS arrays, and different frequency bands to further enhance performance and expand applications.

The study utilized a limited number of participants and activities. The generalizability of the findings to other populations, activities, and environments needs further investigation. The current RIS design is limited to azimuthal beam steering, and future research could explore three-dimensional beamforming. The accuracy gains observed might depend on the specific characteristics of the used RIS, and further optimization could be explored. The study focused on specific scenarios; more extensive testing in diverse real-world environments is necessary to validate its widespread applicability.