Cooperative perception (CP) is attracting increasing attention and is regarded as the core foundation to support cooperative driving automation, a potential key solution to addressing the safety, mobility, and sustainability issues of contemporary transportation systems. However, current research on CP is still at the beginning stages where a systematic problem formulation of CP is still missing, acting as the essential guideline of the system design of a CP system under real-world situations. In this paper, we formulate a universal CP system into an optimization problem and a mobile-edge-cloud framework called Cooperverse. This system addresses CP in a mixed connectivity and automation environment. A Dynamic Feature Sharing (DFS) methodology is introduced to support this CP system under certain constraints and a Random Priority Filtering (RPF) method is proposed to conduct DFS with high performance. Experiments have been conducted based on a high-fidelity CP platform, and the results show that the Cooperverse framework is effective for dynamic node engagement and the proposed DFS methodology can improve system CP performance by 14.5% and the RPF method can reduce the communication cost for mobile nodes by 90% with only 1.7% drop for average precision.

翻译：协同感知（CP）正日益受到关注，被视为支持协同驾驶自动化的核心基础，而协同驾驶自动化是解决当代交通系统安全性、机动性和可持续性问题的潜在关键方案。然而，当前CP研究仍处于起步阶段，缺乏系统性的问题建模——而这一建模正是现实场景下CP系统设计的基本指南。本文提出将通用CP系统建模为优化问题，并构建名为Cooperverse的移动-边缘-云框架。该框架在混合连接与自动化环境中实现CP。引入动态特征共享（DFS）方法来支持特定约束下的CP系统，并提出随机优先级过滤（RPF）方法以高效执行DFS。基于高保真CP平台的实验表明：Cooperverse框架能有效应对动态节点接入，所提出的DFS方法可提升系统CP性能14.5%，而RPF方法在平均精度仅下降1.7%的情况下，将移动节点的通信开销降低90%。