AI-driven Wi-Fi radar people counting for intelligent IoT and distributed sensing systems

Abstract Counting humans in indoor environments is a key capability for intelligent Internet of Things (IoT) systems, enabling applications such as smart home energy optimization, occupancy-aware automation, and real-time decision-making. With the proliferation of connected sensing devices, IoT data analytics must efficiently process large volumes of heterogeneous sensing data while preserving user privacy. In this context, Wi-Fi-based sensing has emerged as a scalable and privacy preserving alternative to vision-based approaches. In this paper, we propose an AI-driven big data analytics framework for device-free people counting using Wi-Fi pulse Doppler radar signals. The system captures spatio-temporal sensing data in the form of range-Doppler images, which are processed using a 3D Convolutional Neural Network (3D-CNN) to extract macro- and micro-movement features. The proposed solution operates under realistic conditions, supporting both Line-of-Sight (LOS) and Non-Line-of-Sight (NLOS) scenarios without constraints on user activity. From a distributed systems perspective, the approach enables real-time processing at the edge using lightweight radar integrated into existing Wi-Fi infrastructure. Experimental results demonstrate high accuracy, reaching 90% in LOS and 94% in NLOS scenarios.