Distributed Cooperative Estimation for Visual Sensor Networks

本项目的目标是为视觉传感器网络建立一个通过分布式信息处理和决策来估计目标运动的智能网络系统，该网络是由多个具有通信和计算能力的智能摄像机组成的网络。本文首先结合前人的两个不同的工作：基于被动的视觉运动观测器和基于被动的姿态同步控制，提出了一种分布式协作视觉估计机制，称为网络视觉运动观测器。该估计机制通过视觉运动观测器执行局部估计，并基于同步技术聚合相邻摄像机的估计。然后，我们基于分布式优化和姿态同步技术，阐明了该机制所产生的估计的平均性能。并对目标运动与平均性能之间的关系进行了理论分析。此外，我们构建了一个实际的视觉传感器网络系统，并验证了所提出的理论结果的有效性。最后，我们量化了恶意传感器和不完美可见性对平均性能的影响，并对传感器网络系统进行了改进，使其即使在存在此类不良事件的情况下也能工作。

The goal of this project is to establish a smart network system estimating target motion via distributed information processing and decision-making for a visual sensor network which is a network consisting of multiple smart cameras with communication and computation capability. We first presented a distributed cooperative visual estimation mechanism, called networked visual motion observer, by combining two different previous works: a passivity-based visual motion observer and passivity-based pose synchronization control. The proposed estimation mechanism executes local estimation via the visual motion observer and aggregates the neighboring cameras estimates based on the synchronization techniques. Then, we clarified an averaging performance of the estimates produced by the proposed mechanism based on the techniques in distributed optimization and pose synchronization. We also conducted theoretical analysis on a relation between target motion and the averaging performance. In addition, we built an actual visual sensor network system and demonstrated validity of the presented theoretical results. Finally, we quantified the effects of malicious sensors and imperfect visibility on the averaging performance and improved the sensor network system so that it works even in the presence of such undesirable events.