Smartphone-based Cooperative Perception
Résumé
We propose a lightweight cooperative perception architecture that leverages smartphones to enhance the situational awareness of non-autonomous vehicles in urban environments. Unlike most existing approaches requiring expensive sensors and dedicated communication modules, our system relies solely on standard smartphone capabilities-namely, the RGB camera and GPS-to detect surrounding vehicles and share their locations with a nearby edge server. The server performs spatio-temporal fusion of the received data and sends back a contextualized, augmented view of the environment to each user.
Implemented and evaluated in the CARLA simulation platform, our solution achieves near real-time performance with an average end-to-end latency of 96 ms per frame and a minimal communication overhead (0.35 KB for uplink, 0.15 KB for downlink). Experimental results show a significant increase in perception coverage and recall, especially for occluded vehicles. This work demonstrates the feasibility of smartphone-based cooperative sensing as a cost-effective alternative to V2X-based systems, with promising prospects for large-scale, infrastructureagnostic deployment.
