NeTS: Small: Distributed Solutions to Smart Camera Networks

智能相机网络融合了计算机视觉、分布式处理和传感器网络等学科，通过分布式感知和协作网络处理提供有价值的信息，从而解决多摄像机应用中的问题。传感器网络中的协作不仅是为了弥补每个传感器节点的处理、感知、能量和带宽限制，而且是为了提高网络的准确性和健壮性。SCN中的协作处理比传统的标量传感器网络(SSN)更具挑战性，因为摄像机具有三个独特的特点，包括极高的数据速率、有限视场(FOV)的定向感知特性和视觉遮挡的存在。开展了一项综合研究，以应对供应链网络带来的独特挑战，其中合作是关键。研究了协同处理的三个方面：1)在存在视觉遮挡的情况下进行覆盖估计，以便在感知覆盖时提供足够的冗余，并在视觉覆盖的统计信息融合了摄像机节点和目标的统计信息的情况下实现协作；2)聚类，通过基于图像比较的语义邻居选择算法在邻居节点之间调度有效的睡眠唤醒模式，以实现高效的协作；3)分布式优化，用于网络数据处理，涉及如何有效地从多个分布式传感器获得稳健而准确的综合结果，以用于在人群中具有挑战性的视觉任务，如目标检测、定位和跟踪。

Smart camera networks (SCNs) merge computer vision, distributedprocessing, and sensor network disciplines to solve problems inmulti-camera applications by providing valuable information throughdistributed sensing and collaborative in-network processing.Collaboration in sensor networks is necessary not only to compensatefor the processing, sensing, energy, and bandwidth limitations of eachsensor node but also to improve the accuracy and robustness of thenetwork. Collaborative processing in SCNs is more challenging than inconventional scalar sensor networks (SSNs) because of three uniquefeatures of cameras, including the extremely higher data rate, thedirectional sensing characteristics with limited field of view (FOV),and the existence of visual occlusion. An integrated research iscarried out to tackle the unique challenges presented by SCNs wherecollaboration is the key. Three aspects of collaborative processingare investigated, 1) coverage estimation in the presence of visualocclusions to provide adequate redundancy in sensing coverage, and toenable collaboration where the statistics of visual coverage blendsthe statistics of camera nodes and targets, 2) clustering toschedule an efficient sleep-wakeup pattern among neighbor nodes formedby image comparison-based semantic neighbor selection algorithm formore efficient collaboration, and 3) distributed optimization, forin-network data processing that concerns how to effectively obtainrobust and accurate integration results from multiple distributedsensors for challenging vision tasks like target detection,localization, and tracking in crowds.