Cooperative Traffic Map Construction in Vehicular Networks

Cooperative perception is fundamental to improving road safety and enabling advanced vehicle applications, where vehicles upload their perception data to the edge server, constructing a comprehensive traffic map that includes all road objects. This traffic map can help drivers be aware of traffic conditions beyond their line-of-sight and make appropriate decisions to avoid potential accidents. However, transmitting and processing large amount of perception data pose significant challenges in vehicular networks with limited bandwidth and computational capacity. Moreover, the GPS sensors might be inaccurate, introducing significant location errors to the traffic map. To address these challenges, we leverage depth images to identify the relevant areas so that only part of the perception data are uploaded and processed, reducing bandwidth consumption and computational burden on the edge server. Then, we construct the traffic map based on the detected objects. Since the same object might be perceived by multiple vehicles, resulting in different detected objects, these corresponding objects should be matched to maintain consistency. We design an efficient algorithm for real-time object matching and improve the location accuracy by transforming relative locations based on the matching. Extensive evaluations demonstrate that our system significantly reduce bandwidth consumption while achieving sub-meter level localization accuracy.