强化学习驱动的基于人脸追踪的远程会议系统
2022 年全国大学生嵌入式芯片和系统设计竞赛
在 2022 年举办的全国大学生嵌入式芯片和系统设计竞赛中，我与 @zobin 组建的团队设计了利用 DDPG 强化学习模型和 YOLOv2 人脸识别模型的远程会议系统。在远程会议的场景下，这个设备可以实现与会人员人脸的实时高精度跟踪。我们的设备使用基于海思 Hi3516 和 Hi3861 芯片的终端平台进行开发，使用 OpenHarmony 作为底层操作系统。我们最终赢得了全国一等奖。
Open Resource as a Service: A More Comprehensive Collaborative Cloud Service Model for Large-scale Heterogeneous Mobile Cloud Computing
6G 网络应用预研项目成果产出
Abstract—In recent years, mobile cloud computing (MCC), as a revolutionary technology, has affected the entire computer industry. With the growth of mobile user number and device categories, more and more researchers are focusing on models that sink services from remote cloud to the proximity of users. As typical representatives, multi-access edge computing (MEC) deploys cloud facilities to base stations, and cloudlet allows edge devices to collaborate. However, large user scale brings a more complicated heterogeneous problem, including openese heterogeneity ignored by previous work. So we propose a new cloud service model, Open Resource as a Service (OpenRaaS), aiming to utilize heterogeneous resources in the network edge. OpenRaaS presents a more open and efficient collaborative method with container technology, which is competent to traditional Software-as-a-Service (SaaS) business and computation offloading services. After analysis and demonstration, OpenRaaS has advantage in compatibility, resource utilization and scalability. We also built an open source cloud desktop-application platform for verification. Under this work, we introduce the concept of Resource-as-a-Service (RaaS), eager to solve the large-scare heterogeneity problem from a more comprehensive view. And we expect OpenRaaS will provide a helpfull insight in service composition and cloud collaboration for future work.
Keywords: Mobile Cloud Computing, Transparent Comput- ing, Cloudlet, Edge Computing, Internet of Things
相关成果正在发表论文与专利。