在靜態場景中追蹤多個物體演算法

Abstract

追蹤物體在智慧型監控系統方面是一個受矚目的議題,如何在發生事 故可以立即得知，並給予及時的幫助。這篇論文提出一個混合式追蹤 物體的方法，用來追蹤物體的資訊包括有物體的輪廓、顏色、移動的 區域性，會分別產生三個的物體相似度，以及會利用我們提出的相對 應的演算法把這三個相似度整合起來進行物體的追蹤。在遮蔽物體方 面，我們儲存物體的輪廓資訊，最後利用這些資訊來作為分離相連物 體的依據。本篇論文可以解決物體追蹤的問題包括剛性和非剛性物體 的出現、消失、分裂、合併、遮蔽現象於場景中。

In the recent years, there have been significant developments in the field of surveillance systems, where object tracking is a key technology. This thesis proposes a new hybrid object tracking method which combines region, edge and location-based methods in the algorithms. For region based method, we use watershed to segment objects into several regions. For edge based method, we use Hough Transform to transform edge from image domain to parameter domain for similarity comparison. The location based method is applied only when both region-based and edge-based methods can’t find the corresponding objects The experimental result shows that the proposed algorithm can track multiple objects in a video sequence with object appearance and disappearance, non-rigid and rigid movements, object splitting and merged as well as object occlusion. The success of tracking rate can be up to 97.9% for video sequence with static scene.