在壓縮格式下監視影片之物體偵測及追蹤

Abstract

對影片中物體的偵測以及追蹤對影片的分類和註解來說,一直是很重要的一環.在這篇論文裡,我們提出了一個自動化偵測並且追蹤物體的系統,能夠針對MPEG格式的影片擷取出物體的移動軌跡.根據MPEG格式中I-Frame跟P-Frame的不同特性,我們分別提出了利用顏色以及移動向量的方法來偵測出在I-frame 以及P-frame中的物體.同時我們也提出了一個利用物體彼此間的空間關係和顏色的組成來比對物體的GOP-based 物體追蹤系統.對影片內容作索引有助於使用者能夠更快速而有效的去瀏覽並且編輯影片片段,因此,我們利用影片中產生的物體移動軌跡來對影片的內容作索引,並且提供了一個完整的軌跡比對系統.利用比對兩兩軌跡的移動方向以及軌跡的長度大小,使用著可以找到他們感到興趣的影片.所有提出來的方法,我們都利用MPEG7的測試資料來作實驗,實驗的結果也都顯示出提出來的方法確實達到有效的目的.

Object segmentation and tracking are the crucial steps for video annotations and classifications in various applications of video surveillance. In this thesis, we propose an automatic object segmentation and object tracking schemes to extract object trajectories in MPEG videos. Background-based object segmentation and motion clustering method is used in I-frames and P-frames respectively for object segmentation. An approach of GOP-based object tracking is proposed, in which spatial and color information are exploited for object matching. Indexing of video contents allows users to effectively and efficiently browse the videos. The moving trajectories produced by object tracking are used as descriptors to represent the video clips. Thus we develop a robust trajectory matching system based on criteria of direction and length of trajectories to provide users an environment to retrieve interested video clips. Proposed methods are evaluated for several MPEG-7 testing sequences. Experimental results demonstrate good performance and thus show the effectiveness of the proposed schemes.