Emp

Accuracy-Aware Cooperative Sensing and Computing for Connected Autonomous Vehicles

“…To enhance the resource efficiency while maintaining the high perception performance, existing works have investigated scalable raw-level cooperative sensing by sharing partial raw sensing data, which allows the transmission and processing of the most relevant segments in the full raw sensing data [15], [16]. For example, a common region of interest (RoI) of multiple CAVs is partitioned to disjoint nonoverlapping spatial areas, and each CAV is responsible for sharing only a segment of raw sensing data for its closest area [15]. As compared with a basic scheme in which all CAVs share the full raw sensing data, the communication and computation resources are greatly reduced without a significant perception accuracy loss.…”

“…With the consideration that data fusion from diverse viewing angles of different CAVs enhances the perception accuracy, the sensing data for each object in the RoI are provided by multiple CAVs in [16]. In the existing works on scalable raw-level cooperative sensing, the principles in partial raw sensing data selection are mainly for performance gain from the localization perspective, e.g., the proximity principle [15] or the relevance of object locations on CAVs' future trajectories [16]. However, it is difficult to estimate the relevance of partial raw sensing data of each CAV from the perspective of improving the object classification accuracy.…”

“…In the conventional broadcast-based cooperative sensing scheme, each CAV locally fuses and processes its own data and all the received sensing data from neighboring CAVs, which incurs large overall computing resource consumption and may lead to delay violation due to the onboard computing resource limitation. To satisfy the stringent perception delay requirement, existing studies focus on the infrastructure support for cooperative sensing of CAVs [15], [17], [18]. By uploading the selected partial raw sensing data from each CAV via vehicle-to-infrastructure (V2I) transmission to an edge server for computation, the more powerful edge computing resources are leveraged, the redundant computation among CAVs for identifying common objects is reduced, and each CAV can obtain the perception results with reduced latency [15].…”

“…To satisfy the stringent perception delay requirement, existing studies focus on the infrastructure support for cooperative sensing of CAVs [15], [17], [18]. By uploading the selected partial raw sensing data from each CAV via vehicle-to-infrastructure (V2I) transmission to an edge server for computation, the more powerful edge computing resources are leveraged, the redundant computation among CAVs for identifying common objects is reduced, and each CAV can obtain the perception results with reduced latency [15]. However, to support the data fusion and processing of selected object sensing data in our fine-grained cooperative sensing scheme, purely relying on a centralized edge server is not the most resource-efficient approach.…”

A novel efficient Multi-view traffic-related object detection framework

“…Briefly, vehicles can share the self-acquired traffic information with other participants (e.g., other vehicles and traffic monitoring equipment). Different from the existing point cloud-based methods [19][20][21][22][23],…”