基於無線感測網路建構最佳節點切換機制進行自動化立體視覺物體追蹤

Abstract

動態物體追蹤是無線感測網路的重要應用之一，由於視覺感測器能夠提供關於觀測物的多樣化資訊，如影像、速度及位置，使其成為該應用中的重要角色。基於無線感測網路的低資料傳輸量限制，如何選擇最佳節點以使用最少的資料頻寬並達到特定應用對資料的品質要求也成為近年來許多研究團隊的議題。過往的研究者發展了許多無線視覺物體追蹤系統，但大多數都只提供物體的平面位置資訊，而其使用二維針孔映射模型定位的方式連帶限制了感測器的擺設。以往針對視覺感測器的所設計的選擇機制也遭遇同樣的問題。 本研究實現一基於立體視覺定位之三維動態物體追蹤系統，並提出三維模糊估測法以進行最佳節點切換。本系統所應用的三維透視投影模型允許高自由度的感測器設置。由於感測器的位置及方向角是視覺定位的重要資訊，本研究同時提供視覺感測器校正程序可同時獲得多感測器間的幾何位置關係。但視覺感測器位置的準確性影響定位結果甚巨，因此本文所提出之最佳節點切換機制將感測器本身位置誤差視為選擇時需考量的要素之一。 本系統經過模擬及實驗的驗證，其靜態定位誤差與另兩種在三維空間中可實施切換機制:最近感測器組及隨機選擇相比皆有顯著性的差異。在方形與曲線動態追蹤測試中也與整合多攝影機資料的定位誤差相近或甚至與之相比更為準確。改變視覺感測器之高度與垂直旋轉角度同樣能有良好的定位結果，證明此節點切換機制於不同的感測器擺設下之適用性。在方形路徑追蹤過程中，平均有31.3 %的封包節省率，而在曲線路徑中有49.8%的封包節省率，該數據說明本文所提出的動態物體追蹤機制達到了低資料量並具準確性的定位結果需求。本最佳節點切換機制之動態物體追蹤系統未來可結合長時間運行之安全監視或居家看護系統以提供高可靠性之人員或物體之位置資訊。

Object tracking is one of the appealing applications in wireless sensor network. Vision sensors can be used to provide vital information on the objects for tracking problem. The major limitation in wireless sensor network is low bandwidth for data transmission, in order to balance the trade-off between the limitation and the service quality, the sensors with most informative measurements should be selected properly. The previous proposed vision object tracking systems and sensor switching schemes which utilized 2 dimension pin-hole model only provide the position information in certain plane and restrict the sensor configuration. This study designed a vision object tracking system with optimal sensor switching mechanism based on 3 dimension uncertainty measurement. The stereo vision localization technique which is utilized in this study does not set the limitation on the sensor configurations. Since the position and orientation of the vision sensor are essential for stereo localization, a calibration process is implemented for obtaining the geometry information of the sensors. However, calibration error will diminish the localization accuracy, the proposed sensor switching mechanism takes this error as an impact factor which didn’t mentioned explicitly in previous research. The proposed object tracking system and sensor switching mechanism were verified and evaluated through simulations and experiments. There is significant difference in the static localization error between the proposed switching mechanism and other two selection policy which is able to apply in 3 dimension space: closest set and random selection. The proposed switching mechanism caused average 31.3 % packet reducing in square route tracking and 49.8 % in curve route from practical implementation. It also gave the dynamic tracking error close and even is lower the result from combination of multiple sensor data. The outcome proved that the tracking system achieved the goal of data redundancy and maintains reasonable tracking error by switching sensor set dynamically. The tracking system proposed in this study has apparent potential to merge with long-time operating application such as secure surveillance monitoring and home telecare systems.