在壓縮格式下一個以紋路與空間資訊來加強位移向量之創新架構

Abstract

在這篇論文裡，我們提出了一個創新的方法，針對MPEG-2的影像資料串流以健全完整的位移向量為基礎做物件偵測。我們對擷取出的位移向量，使用後處理的方式來取代直接自壓縮格式中取出的位移向量， 以減少雜訊影響及獲得更健全完整的物件資訊。透過我們提出包含空間資訊過濾元件、時間資訊過濾元件以及紋路過濾元件之系統， 我們可以使這些資訊變得更精確。基於我們提出的架構，物件偵測演算法能更正確地偵測物件，並得到更快的處理效能。 我們利用MPEG-7的測試資料作為我們提出的系統與其他受歡迎和廣泛被使用的方法效能比較之實驗輸入。由實驗結果，我們的系統明顯的比其他方法來得好。此外，我們也提出了一個互動式參數調整使用者介面。

In this thesis we propose a novel approach for robust motion vector based object detection in MPEG-2 video streams. Rather than processing the extracted motion vector fields that are directly extracted from MPEG-2 video streams in the compressed domain, we perform post processing operation over the extracted motion vector, in order to reduce the noise within the motion vector content, and to obtain more robust object information. We refine this information through our proposed system which composed of a Spatial filter Component, a Temporal filter Component and a Texture filter component. As a result, the object detection algorithm is more capable of accurately detecting objects with more efficient performance in terms of runtime. We compare the performance of our proposed system with other popular and commonly related work and techniques. Based on the experimental results performed over the MPEG7 testing dataset and measuring performance by using the standard recall and precision metrics, object detection using our proposed system is remarkably superior to the alternative techniques. In addition to these results, we describe a user system interface that we developed, where users can maintain the parameters interactively.