Cooperative Perception for Connected Autonomous Vehicles Under Constrained V2V Networking
Cooperative Perception for Connected Autonomous Vehicles Under Constrained V2V Networking
复制标题
DOI：
10.1109/ieeeconf59524.2023.10476810
复制
发表时间：
2023-10
期刊：
影响因子：
--
通讯作者：
Ahmad Sarlak;Sayed Pedram Haeri Boroujeni;Hazim Alzorgan;Rahul Amin;Abolfazl Razi;Hossein Rajoli
中科院分区：
文献类型：
--
作者：
Ahmad Sarlak;Sayed Pedram Haeri Boroujeni;Hazim Alzorgan;Rahul Amin;Abolfazl Razi;Hossein Rajoli
协作感知 (CP) 是一种新兴技术，可用于增强自动驾驶汽车 (AV) 的安全性，因为仅依靠自动驾驶汽车自身的摄像头可能无法产生准确的结果。此类条件的示例包括雾天、蜿蜒道路以及前方车辆遮挡摄像头视野下的物体检测延迟或精度低。最近提出了一些 CP 方法来通过协作感知来提高 AV 感知的质量。尽管取得了成功，但这些方法仍存在一些局限性，例如对完美通信和不受约束的资源以及无法访问大型训练数据集采取不切实际的假设。在本文中，我们使用 LTE Release 14 模式 4 侧链路通信通过选择性 V2V 通信来实现 CP，以实现态势感知。我们的方法基于需求响应机制，并解决使用辅助车辆的优化问题，考虑网络性能指标（例如丢包率）、可用资源（LTE-V 中的无线电块）、扩展视觉范围（覆盖的总路段），以及更重要的是所选车辆的最终感知质量。我们的初步结果表明，与传统的单摄像头视觉相比，所提出的方法具有显着的增益。
Cooperative Perception (CP) is a newly emerged technique that can be used to enhance the safety of Autonomous Vehicles (AVs) when relying merely on the AV's own camera may not yield accurate results. Examples of such conditions are delayed or low object detection accuracy under foggy weather, winding roads, and blocked camera vision by the front vehicle. A few CP methods have been recently proposed to enhance the quality of AV perception through cooperative perception. Despite their success, these methods have a few limitations, such as adopting unrealistic assumptions about perfect communication and unconstrained resources as well as having access to large training datasets. In this paper, we use the LTE Release 14 Mode 4 side-link communication to implement CP by selective V2V communication for situational awareness. Our method is based on a demand-response mechanism and solving an optimization problem to employ helper vehicles, considering networking performance metrics (such as packet drop rate), available resources (radio blocks in LTE-V), extended visual range (the total road segment covered), and more importantly the ultimate perception quality of the selected vehicles. Our preliminary results demonstrate a significant gain for the proposed method compared to a conventional single-camera vision.