面向车联网自动驾驶的边缘智能多源数据处理

在车联网场景中，智能车辆受限于自身有限的感知范围和计算资源，难以实现高可靠性的自动驾驶.针对上述问题，提出了面向车联网自动驾驶的边缘智能多源数据处理方法，通过发掘网络中节点的空闲资源，提高系统的吞吐量，利用网络中的分布式数据，提高神经网络的推断准确率.为了解决车辆节点数据传输的耦合问题，构建了3层网络模型，对网络中的数据传输进行调度，并将原车辆节点一一映射为一个虚拟节点集合，对节点的数据传输进行解耦，从而把原问题转化为最小费用最大流的问题，实现了对自动驾驶网络计算、通信资源的高效分配.仿真结果表明，相比于本地处理的方案，所提方案可提升150%的系统吞吐量和12%的系统能效.

Edge Intelligence Multi-Source Data Processing for Autonomous Driving in Internet of Vehicles

Designing a highly reliable automatic driving system is challenging due to the limited perception range and computing resources of intelligent vehicles. To this end, the distributed data and available resources in the network should be exploited. An edge intelligence multi-source data processing scheme for autonomous driving in Internet of vehicles is proposed. Firstly, a three-layer network model is constructed to schedule the data transmission. Then the original vehicle nodes are mapped one by one into a set of virtual nodes to decouple the data transmission. Thereafter, the original problem into a minimum cost maximum flow problem, realizing the efficient allocation of computing, communication resources in the network are transformed. Simulations show that the proposed scheme can improve system throughput by 150% and system energy efficiency by 12% compared with local processing scheme.