- 學位論文
智慧型系統在移動物體辨識上之應用
The Application of Intelligent System to the Moving Object Recognition
摘要
近來監控系統廣被接受，多數的監控系統是利用數位影像儲存裝置，儲存監視器所拍攝到的畫面，而近來的應用朝向當有物體移動，監視器便開始錄影的方向發展。本論文提出使用模糊推論、倒傳遞網路以及灰預測理論建立一套自動監視系統。系統開始會先讀取由監視器拍攝的影像，接著使用移動物體偵測技術，在獲取移動物體的訊息後，再利用上述的理論以完成人形辨識、行人身高計算、行人未來路徑預測以及機車騎士身高的計算。在人形辨識方面，本論文使用模糊推論來辨別場景中的物體是人或是其它動物，利用人與動物軀幹的不同，我們將軀幹分成兩組，一組是表示人的軀幹，另一組表示動物的軀幹，每一組都建有四種模型。本系統會將偵測的物體與這八種模型做相似度的計算，由相似度值判斷該物體是屬於人或者是屬於動物。若本系統判斷該物體為人，接著就計算其實際身高。這部分我們採用類神經網絡中最為人所熟知的倒傳遞網路，因為每個監視器所架設的地方不同，因此利用倒傳遞網路訓練一組模組，以適合其架設之地方，待將數據輸入該模組後，可產生該物體實際身高值。 當監視器完成擷取四個畫面後，本系統再利用灰預測中的GM(1,1)模型，以預測該物體在畫面中未來移動的路徑。在發現該物體闖進預設的警戒區時，本系統會即刻發出警報，以預防外物的入侵。本論文除有完整的理論探討外亦有詳細的模擬結果來驗証所提方法之有效性。
關鍵字
智慧型系統並列摘要
The monitoring system has been widely accepted recently. Most monitoring systems utilize the digital image storage device to store the captured pictures. The recent application focuses on recording those moving objects only. In the thesis, we propose an automatic surveillance system which employs the fuzzy inference, Back Propagation Network (BPN) and grey prediction theory. First, the system will read the image captured from the charge coupled device (CCD), and then we utilize the object detection technology to analyze the moving object. After deriving the moving object information, the grey prediction model is employed to process the human recognition, human’s height calculation, human’s moving path prediction and motorcyclist’s height calculation. Due to the human’s skeleton is different from that of the animal, we can utilize the fuzzy inference to differentiate the moving objects belonging to human or animals. Therefore, we divide the skeleton into two categories: the human and the animal. Four skeleton models are built for human and animals, respectively. These models are sufficient to calculate the similarity of moving objects within an image. After recognizing a human, the system utilizes the well-known artificial neural network, back propagation network, to calculate the real height of that object. This information can help identify the moving objects. Followed by coupling with the GM(1,1) model, our system can predict the object future path. Based on the object moving path, the system can make an early-warning in case someone breaks into restricted area. Not only the theoretical study is studied, but also detailed simulation results are presented to verify the effectiveness of the proposed system.