COSMOS Smart Intersection: Edge Compute and Communications for Bird's Eye Object Tracking
COSMOS Smart Intersection: Edge Compute and Communications for Bird's Eye Object Tracking
复制标题
DOI：
10.1109/percomworkshops48775.2020.9156225
复制
发表时间：
2020-03
期刊：
影响因子：
--
通讯作者：
Shi-hua Yang;Emily Bailey;Zhengye Yang;J. Ostrometzky;G. Zussman;I. Seskar;Z. Kostić
中科院分区：
文献类型：
--
作者：
Shi-hua Yang;Emily Bailey;Zhengye Yang;J. Ostrometzky;G. Zussman;I. Seskar;Z. Kostić
智能城市交叉口将在未来城市的自动交通管理和行人安全改善中发挥至关重要的作用。它们将（i）聚合来自车载和基础设施传感器的数据;（ii）利用低延迟高带宽通信、边缘云计算和基于AI的物体检测和跟踪来处理数据;（iii）为控制系统提供智能反馈和输入。用于城市规模部署的云增强开放式软件定义的移动的无线测试床（COSMOS）支持对支持智慧城市的技术进行研究。在本文中，我们提供的实验结果，使用鸟瞰相机检测和跟踪车辆和行人的COSMOS试点。我们通过评估和定制一系列视频预处理和深度学习算法来评估实时计算、检测和跟踪精度的能力。探索并解决了与行人与汽车鸟瞰图的比例差异相关的不同问题：汽车的最佳多对象跟踪精度（MOTA）约为73.2，行人约为2.8。每秒30帧的实时目标-即，一旦处理时间被大致改进三倍，则车辆的对象检测的总共33.3ms的等待时间将是可达到的。
Smart-city intersections will play a crucial role in automated traffic management and improvement in pedestrian safety in cities of the future. They will (i) aggregate data from in-vehicle and infrastructure sensors; (ii) process the data by taking advantage of low-latency high-bandwidth communications, edge-cloud computing, and AI-based detection and tracking of objects; and (iii) provide intelligent feedback and input to control systems. The Cloud Enhanced Open Software Defined Mobile Wireless Testbed for City-Scale Deployment (COSMOS) enables research on technologies supporting smart cities. In this paper, we provide results of experiments using bird's eye cameras to detect and track vehicles and pedestrians from the COSMOS pilot site. We assess the capabilities for real-time computation, and detection and tracking accuracy - by evaluating and customizing a selection of video pre-processing and deep-learning algorithms. Distinct issues that are associated with the difference in scale for bird's eye view of pedestrians vs. cars are explored and addressed: the best multiple-object tracking accuracies (MOTA) for cars are around 73.2, and around 2.8 for pedestrians. The real-time goal of 30 frames-per-second - i.e., a total of 33.3 ms of latency for object detection for vehicles will be reachable once the processing time is improved roughly by a factor of three.