AI-based Wi-Fi sensing used for human presence detection and activities classification, by using the characteristic of the wireless channel, such as Channel State Information (CSI), to implement a device-free sensing system inside vehicles. Compared to conventional in-cabin sensing technologies such as cameras, pressure mats, and dedicated radar or infrared sensors, Wi-Fi sensing can preserve more privacy solution, operate reliably under challenging lighting conditions, and be integrated with relatively low additional hardware complexity. A reliable in vehicle sensing system is increasingly important for safety and comfort applications such as occupant presence and seat detection, driver monitoring, and child presence awareness.
Despite promising accuracy reported in controlled demonstrations, the performance of AI-based Wi-Fi sensing in vehicles is still highly sensitive to environmental factors that come from the sensitivity of the electromagnetic signals to environmental changes. Vehicular environments technically act as a small indoor environment which is suitable for WiFi sensing. The vehicular environment in general faces some challenges when it comes to applying WiFi sensing, such as fluctuations in the cabin’s temperature, sunlight exposure and air-conditioning operation, Wi-Fi transceiver type and placement, and wireless configuration parameters such as channel bandwidth. These factors affect multipath propagation, hardware behavior, and noise characteristics, often leading to significant performance degradation when AI models are deployed outside their training conditions.
This project addresses the lack of systematic analysis of such variability by investigating the impact of key in-cabin environmental and operational factors on the reliability and generalization of AI-based in-vehicle Wi-Fi sensing. This research will be focused on collecting a comprehensive multi-condition in-vehicle Wi-Fi sensing dataset under environmental and operational variability to study the effect of such conditions on the CSI signal. The project will further develop signal preprocessing and feature engineering pipelines, implement and benchmark multiple AI-based sensing models, and systematically assess performance degradation and failure modes across conditions and their interactions. The outcomes of the project are expected to provide critical insights, benchmarks, and design guidelines that support dependable deployment of AI-based Wi-Fi sensing in real vehicular environments.