VESTA: In-Vehicle Sensing with GNSS Satellite Signals

Abstract

Vehicles have become an indispensable part of modern life. Over the past decade, perception systems such as radar and LiDAR have become increasingly popular for out-vehicle sensing. In contrast, comparatively little attention has been paid to in-vehicle sensing. While camera-based solutions raise privacy concerns, wireless sensing-based approaches typically require dedicated transmitters, increasing deployment costs and potentially interfering with existing wireless communications. In this paper, we propose VESTA, an in-vehicle sensing system that leverages the ubiquity of GNSS signals. The ubiquitous availability of satellite signals, together with commodity GNSS receivers embedded in smartphones and vehicles, makes in-vehicle sensing feasible without requiring dedicated RF transmitters. To realize this system, we leverage a vehicle-mounted reference GNSS receiver to compensate for vehicle-induced motion interference. We further develop a novel multi-satellite signal processing scheme that enables angle-of-arrival (AoA) estimation using a single-antenna GNSS receiver, and design a lightweight fusion network that incorporates satellite geometry to address the unique challenges of the in-cabin sensing environment. In on-road experiments covering over 1,000 km, VESTA achieves higher than 90% recognition accuracy with end-to-end latency below 100 ms across diverse vehicles, roads, and weather conditions.