Attention-Enhanced Deep Learning for Device-Free Through-the-Wall Presence Detection Using Indoor WiFi Systems

TL;DR

This paper addresses indoor device-free presence detection using WiFi channel state information (CSI) in through-the-wall settings. It introduces ALPD, which combines an attention-based subcarrier weighting scheme with a bidirectional CNN-LSTM feature extractor, and employs a Subcarrier Cluster Autoencoder (SCAE) to select informative subcarriers. The offline training integrates subcarrier selection, time-domain feature extraction, and feature mapping across two transmission pairs, while online prediction delivers real-time presence decisions. Experiments in a two-room through-the-wall scenario show ALPD consistently outperforms SVM, FCN, and NC-LSTM benchmarks, remains robust under interference, and benefits from bidirectional transmission and static/dynamic feature fusion, achieving near 96% accuracy.

Abstract

Accurate detection of human presence in indoor environments is important for various applications, such as energy management and security. In this paper, we propose a novel system for human presence detection using the channel state information (CSI) of WiFi signals. Our system named attention-enhanced deep learning for presence detection (ALPD) employs an attention mechanism to automatically select informative subcarriers from the CSI data and a bidirectional long short-term memory (LSTM) network to capture temporal dependencies in CSI. Additionally, we utilize a static feature to improve the accuracy of human presence detection in static states. We evaluate the proposed ALPD system by deploying a pair of WiFi access points (APs) for collecting CSI dataset, which is further compared with several benchmarks. The results demonstrate that our ALPD system outperforms the benchmarks in terms of accuracy, especially in the presence of interference. Moreover, bidirectional transmission data is beneficial to training improving stability and accuracy, as well as reducing the costs of data collection for training. To elaborate a little further, we have also evaluated the potential of ALPD for detecting more challenging human activities in multi-rooms. Overall, our proposed ALPD system shows promising results for human presence detection using WiFi CSI signals.

Figures (14)

• Figure 1: The scenario of through-the-wall human presence detection. Two rooms are deployed with an AP, where each AP can be operated as either Tx or Rx at one time controlled by the server. The CSI signal includes LoS and NLoS due to multipath effects, which are collected by AP and processed by the server.
• Figure 2: Preliminary experimental observations on CSI amplitude versus subcarrier index from four antenna pairs: An empty room case with CSI (a) before and (b) after normalization. The remaining presence cases are collected with normalized CSI, including a person randomly walking in the room deployed with either (c) Tx or (d) Rx, as well as a person standstill at either (e) LoS or (f) NLoS.
• Figure 3: Schematic diagram of the proposed ALPD system.
• Figure 4: Architecture of attention-based subcarrier selection.
• Figure 5: Architecture of time-domain feature extraction. Note that three-layer CNN and two-layer LSTM are considered for extracting spatial and temporal features from CSI signals.