於視覺遮蔽環境下針對地面目標之空中協同軌跡追蹤定位與監視系統

Abstract

無人機可用於許多空中任務，例如探索、檢查、建圖、與環境互動與搜救。在這些無人機的應用中，其中一項是針對警隊。在市區中，追捕嫌疑人或目標是市區警隊的重要任務之一，此方式無需消耗過多人力。然而，長期的訓練過程是非常勞力密集的。同時，在任務過程中，一些畫面的目標估測失誤可能也會發生，這可能是由於障礙物或訊號傳輸故障造成的遮擋情形引起的。 因此，在這篇論文中，我們使用一四旋翼無人機隊解決針對單一目標之空中多群體自主追蹤問題，而此無人機隊會受到障礙物或無法預測之暫時訊號傳輸故障遮擋的影響，造成目標估測暫時失靈。此系統包含使用卡爾曼濾波器(Kalman Filter)融合機載圖像與慣性測量單元(IMU)測量數據進行無人機的全局定位，根據與無人機相對位置估測目標位置，透過基於軌跡之動作預測解決目標位置暫時的目標估測失效，提取唯一目標狀態的平均演算法，以及基於共識(Consensus)之多群體系統隊形追蹤控制演算法。此外也提供了實驗結果來展示系統的可用性。

UAVs (Unmanned Aerial Vehicles) or drones can be used for many purposes in a lot of aerial tasks, such as exploration, inspection, mapping, interaction with the environment, or even search and rescue. Among these applications of the UAVs, one of them is the police team. Chasing suspects or targets can be one of the most important task for the police in the city without much effort. However, the long process of training would be labor-intensive. Simultaneously, during the tasks, there might be some image-based estimation failure of the target, which may be caused by occlusion from the obstacles or signal transmission failure. As a result, in this thesis, we solve the autonomous aerial multi-agent tracking problem toward a single target using a team of quadrotors under the occlusion by obstacles or unpredicted temporary signal transmission failure. The system include global localization of quadrotors using data fusion of onboard image and IMU measurement by Kalman filter (KF), target position estimation with the positions relative to the quadrotors, trajectory-based target motion prediction for solving temporarily unpredictable estimation failure of the target position, averaging algorithm for extracting the unique state of the target and consensus-based formation tracking algorithm for controlling the multi-agent system. Furthermore, We also provide some experiment results to demonstrate the capability of the system.