TL;DR
This paper introduces an adaptive, real-time indoor intrusion detection system that leverages Channel State Information (CSI) from commodity Wi-Fi devices. By analyzing the correlation patterns of OFDM subcarriers, the system achieves high accuracy in detecting intrusions across various challenging scenarios, including different intruder speeds, multiple intruders, and even when occupants are sleeping.
What Problem Does This Paper Address?
Traditional indoor intrusion detection systems (IDS) often rely on dedicated sensors (like infrared or ultrasonic) or cameras. However, these systems have limitations: dedicated sensors can be intrusive and prone to environmental drift, while cameras raise privacy concerns and are sensitive to lighting conditions. Device-free sensing using Wi-Fi signals offers a promising alternative, but existing Wi-Fi-based IDSs struggle with adaptability to diverse real-world environments. They often fail to account for variations in intruder motion speed, the number of intruders, non-line-of-sight (NLOS) conditions, and the presence of sleeping occupants, leading to performance degradation. This paper aims to bridge this gap by developing an IDS that is robust and adaptive to these complex scenarios.
Methods
The core of the proposed system lies in its novel feature extraction and selection approach based on the empirical analysis of OFDM subcarriers.
- OFDM Subcarrier Analysis: The system exploits the characteristics of narrow frequency spacing between adjacent OFDM subcarriers. It's observed that the Doppler effect, caused by moving targets, manifests similarly across these closely spaced subcarriers.
- Link Pair Selection: Instead of randomly selecting Wi-Fi links or averaging all subcarrier data, the authors propose an algorithm to select an optimal link pair. This selection is based on minimizing the variance of the correlation coefficient matrix entries, aiming to find the most sensitive link for detecting human motion.
- Feature Extraction: Two key features are extracted:
- Detail Features: The sum of correlation coefficients of adjacent subcarriers.
-
General Features: The sum of correlation coefficients of all subcarriers exceeding a certain threshold (ρ). These features are derived from the correlation coefficient matrix of subcarriers, which is shown to be more stable and less affected by environmental factors than raw amplitude or variance.
-
Preprocessing: The system incorporates several preprocessing steps to enhance robustness:
- Outlier Detection: Local Outlier Factor (LOF) is used to identify and remove abnormal CSI data points.
- Wavelet-Based Filtering: Discrete Wavelet Transformation (DWT) is employed for de-noising, effectively removing noise while preserving crucial human motion information.
- Normalization: Min-max normalization is applied to standardize CSI amplitudes to a [-1, 1] range, making them comparable across different environments.
- Classification: A Support Vector Machine (SVM) with a Radial Basis Function (RBF) kernel is used as the classifier. The system is trained offline and then deployed for real-time detection.
Datasets
The paper describes experiments conducted in a real apartment setting, simulating various intrusion scenarios. While no specific public datasets are mentioned for training or evaluation, the authors collected their own data by setting up a Wi-Fi router (AP) and a laptop with a compatible NIC (MP) in a real apartment. They designed specific scenarios including:
- Line-of-Sight (LOS) and Non-Line-of-Sight (NLOS) conditions.
- Different intruder speeds (slow and fast).
- Varying numbers of intruders (single and multiple).
- Scenarios with occupants sleeping (including turning over and normal breathing).
This approach of using a real-world environment and creating diverse, challenging scenarios differentiates their work from many existing studies that often rely on laboratory settings.
Key Insights and Ideas
- OFDM Subcarrier Correlation is Key: The paper highlights that the correlation patterns between adjacent OFDM subcarriers are highly sensitive to human motion and can effectively distinguish between static and dynamic states, even in complex environments. This is a more robust feature than relying solely on amplitude or variance.
- Optimal Link Pair Selection is Crucial: Randomly picking Wi-Fi links or averaging all subcarrier data is unreliable. Identifying the most sensitive link pair significantly improves detection performance.
- Adaptability to Real-World Scenarios: The system's ability to perform well under challenging conditions like NLOS, multiple intruders, and sleeping occupants is a significant advancement over many existing Wi-Fi sensing systems.
- Sleeping Occupant Detection: The system demonstrates a remarkable ability to detect intrusions even when occupants are sleeping, a scenario often overlooked in prior research. This is achieved by incorporating body turning and breathing information into the static data.
- Speed-Sensitive Features: The chosen features are sensitive enough to differentiate between low and high intruder speeds, which is valuable for more nuanced intrusion detection.
Key Results
- High Overall Accuracy: The system achieves an impressive overall precision of 98.96% and a low false negative rate (FNR) of 0.73% under LOS conditions.
- Robustness in NLOS: Even in NLOS environments, the system maintains high precision (98.17%) with a slightly increased FNR of 3.32%.
- Performance Across Scenarios:
- Speed: High precision (97.94% for low speed, 99.09% for high speed).
- Number of Intruders: High precision (97.53% for single, 99.02% for multiple).
- Sleeping: Precision of 97.52% and FNR of 1.45% in sleeping scenarios.
- Comparison with State-of-the-Art: The proposed method significantly outperforms eigenvalue-based and threshold-based methods in terms of precision and FNR across various scenarios, especially in the challenging NLOS environment. For example, in NLOS, the proposed method achieves 98.17% precision compared to 70.51% for the threshold-based method.
- Real-time Performance: The system can detect intrusions within 2 seconds with a sliding window size of 2s and a step size of 0.1s, making it suitable for real-time applications.
Limitations and Future Work
- Limitations:
- The paper acknowledges that while the system performs well, the phase information of CSI was not utilized due to calibration complexity and bandwidth concerns. This could potentially offer further improvements.
- The experimental setup, while realistic, was conducted in a controlled environment within a real apartment. Further validation in more diverse and uncontrolled real-world settings might be necessary.
- Future Work:
- Phase Information: Incorporating phase information of CSI for enhanced detection capabilities.
- Wi-Fi 6 Support: Extending the system to support the next generation Wi-Fi 6 (IEEE 802.11ax) standard.
- Speed Detection: Leveraging the system's sensitivity to motion speed for dedicated Wi-Fi-based speed detection applications.
- Crowd Counting: Exploring the potential for crowd counting using the developed techniques.
Summary updated