DroneNet-Sim: a learning-based trace simulation framework for control networking in drone video analytics
DroneNet-Sim: a learning-based trace simulation framework for control networking in drone video analytics
复制标题
DroneNet-Sim：基于学习的跟踪模拟框架，用于无人机视频分析中的控制网络
DOI：
--
复制
发表时间：
2020
期刊：
影响因子：
--
通讯作者：
P. Calyam
中科院分区：
文献类型：
--
作者：
Chengyi Qu;Alicia Esquivel Morel;Drew Dahlquist;P. Calyam
配备摄像头的无人机（UAV）或无人机广泛用于不同的环境态势感知应用，例如：智能农业，边境安全，智能交通。用于开发与视频流/分析相关的新型网络控制算法的真实无人机测试床构建是耗时且困难的。在执行大规模无人机视频分析实验时会遇到挑战，原因包括：无人机制造的限制、政府监管限制以及有限的能源资源。此外，开发算法需要能够理解网络协议选择的影响，以处理不同的UAV移动模型以及边缘网络视频处理期间的动态网络状态。在本文中，我们提出了一种新的基于学习的跟踪仿真框架，即“DroneNet-Sim”，它集成了无人机和网络侧的仿真。它允许对网络协议选择进行实验（即，TCP/HTTP、UDP/RTP、QUIC）和视频属性选择（即，编解码器、分辨率），以确保在不同的无人机飞行场景中提供令人满意的视频质量。使用机器学习模型，我们展示了DroneNet-Sim如何处理真实世界的无人机轨迹，其中包括各种移动模型，地理空间链接信息和从真实世界的数据收集工作中获得的实时网络状态。使用我们的DroneNet-Sim进行的基于跟踪的实验显示了视频质量交付（即，PSNR）在机器学习模型精度方面与真实世界的测量结果相匹配。
Unmanned Aerial Vehicles (UAVs) or drones equipped with cameras are extensively used in different applications for environmental situational awareness such as: smart agriculture, border security, intelligent transportation. Realistic UAV testbed building for developing novel network control algorithms relating to video streaming/analytics is time-consuming and difficult. Challenges arise when executing high-scale drone video analytics experiments due to: constraints in drone manufacturing, government regulation restrictions, and limited energy resources. Also, developing algorithms requires ability to understand impact of network protocol selection to handle diverse UAV mobility models as well as dynamic network status during edge-network video processing. In this paper, we propose a novel learning-based trace simulation framework viz. "DroneNet-Sim" that integrates simulation on both drone and networking sides. It allows for experimentation with network protocol selection (i.e., TCP/HTTP, UDP/RTP, QUIC) and video properties selection (i.e., codec, resolution) to ensure satisfactory video quality delivery in different drone flight scenarios. Using machine learning models, we show how DroneNet-Sim can process real-world drone traces that include various mobility models, geospatial link information and on-time network status obtained from real-world data-gathering efforts. Trace-based experiments with our DroneNet-Sim shows how video quality delivery (i.e., PSNR) using suitable control networking matches real-world measurements in terms of machine learning model accuracy.
DOI：
10.1109/icnp.2019.8888089
发表时间：
2019-10
期刊：
2019 IEEE 27th International Conference on Network Protocols (ICNP)
影响因子：
--
作者：
Chengyi Qu;Songjie Wang;P. Calyam
通讯作者：
Chengyi Qu;Songjie Wang;P. Calyam
无人机视频分析中基于遮挡检测的动态计算卸载和控制
DOI：
10.1145/3369740.3369793
发表时间：
2020
期刊：
ICDCN 2020: Proceedings of the 21st International Conference on Distributed Computing and Networking
影响因子：
--
作者：
Ramisetty, Rajeswara Rao;Qu, Chengyi;Aktar, Rumana;Wang, Songjie;Calyam, Prasad;Palaniappan, Kannappan
通讯作者：
Palaniappan, Kannappan