安全監控系統中遺留物及其遺留者偵測之研究

Abstract

在公共安全監控領域，關於遺留物檢測及遺留者偵測的議題越來越多，有心人士惡意遺留危險物品或者鐵軌、公路車道之障礙物皆會威脅到社會安全，另外，也常有行人不小心遺留了物品於公共環境，針對上述事件，目前大多使用人員進行監控及事後一一檢閱大量影片進行搜證，十分耗費人力，而且效率不高。本論文針對上述問題，設計一套對遺留物及遺留者偵測方面有所幫助的智慧型監控系統，主要目的在於減少人力成本，免除人員為找出可疑畫面而逐一檢索所耗費大量的時間，及提高監控系統的安全性與可靠性。 在此，本論文的貢獻有以下三點：第一，本論文提出一組特徵，其包含幾何不變矩及方向梯度直方圖特徵，使得在遺留物偵測的執行上更準確及有效；第二，偵測到遺留物之後，透過警戒範圍篩選及追蹤可疑人物，並提出一套投影量分析機制及角度估測計算，可以改善誤判非遺留者為遺留者之情況；第三，完成一個遺留物及其遺留者之自動偵測系統雛形，並於若干不同公共環境下進行驗證，證實本系統能有效的減輕原先逐一檢索畫面所需之人力與縮短其所耗費的時間。

In the field of public security surveillance system, the issues about abandoned objects and abandoner detection are more and more popular. The society security will be threatened by the dangerous objects abandoned intentionally as well as obstacles on the way of bus and railway. Besides, people often leave something in the public region accidentally. For the aforementioned events, institution checks the videos by people to find the abandoned objects and abandoner after events happen. In this thesis, we design an intelligent video surveillance to reduce the loading of human resource. There are three contributions of this thesis. First, this thesis proposes a set of features, including the invariant moment and histogram of gradients to detect the abandoned object accurately. Second, this thesis proposes an analysis of projection vector and estimation of angle to improve the accuracy of the abandoner detection. Finally, this thesis completes the prototype of system in abandoned objects and their own abandoner detection to reduce the loading of employees.