Advanced Wireless Multitarget Activity Detection Using Harmonic Beam Generation and Convolutional Neural Networks

The increasing demand for efficient and scalable human activity recognition (HAR) systems, especially in crowded environments, has highlighted the limitations of traditional methods. These systems often struggle with signal interference and overlapping targets, affecting their precision and scalability. This study presents a wireless system designed for multitarget HAR, leveraging advancements in spatiotemporal coding reflecting surface. By utilizing a programmable reflecting surface, the system generates harmonic beams that enhance angular resolution, allowing for the detection and classification of multiple activities from individuals in close proximity. The reflecting surface’s ability to manipulate the propagation of electromagnetic waves creates independent detection channels, significantly improving signal separation. The system operates using a single-input single-output (SISO) configuration, reducing hardware complexity while maintaining high performance. In addition, digital down-conversion (DDC) is employed to isolate harmonic frequency components, ensuring precise activity recognition. AI-based classification algorithms process the received signals to identify activities such as walking, sitting, and exercising. Experimental results demonstrate superior classification accuracy, offering a scalable and high-resolution solution for real-time human activity monitoring. This innovative approach has potential applications in healthcare, surveillance, and smart home systems, where efficient and reliable activity detection is critical.