標題: 在影帶中基於粒子濾波器的多重部分人體追踪
Particle filter-based Multi-part Human Tracking in Video Sequences
作者: 李金翰
Li Jin-Han
李錫堅
Hsi-Jian Lee
資訊科學與工程研究所
關鍵字: 粒子濾波器;追蹤;物體偵測;Particle Filter;Tracking;Object Detection
公開日期: 2006
摘要: 本論文之研究目的在於如何利用現有的技術再加強去追踪人體。在一個監控環境中,我們希望建立一個機制去追踪影帶中的人體進而利用相對資訊來做應用。隨著時代的進步,使得視訊訊號數位化處理更為普及化,因此使得智慧型視訊監控系統(VASM)更為矚目。也因為此系統更能切合大眾的需求,因此智慧型監控系統在居家安全控管上扮演很重要的角色。 以往在傳統上為偵測入侵者必須要人工以手動的方式一一監看監視系統,因此我們希望可以配合影像處理以及物體追踪技術,以自動化的方式提供智慧型的監控系統,對居家安全有更一層的保障。 在提出的系統中,分成四個部分: 偵測出影像中的移動物體,對前景物體分割成三區塊並同時分別以三區塊做追蹤的動作,利用三區塊的資訊去判斷異常,利用異常的資訊調整區塊 : 第一個部分,影帶中前景人物的偵測。由於我們的系統目標是希望能夠更加有效的進行人體追蹤,故在前景偵測的部分我們已假設在追蹤過程前都已經明確知道,如果便是配合我們前景偵測的技術去找出影帶中的移動人體,進而利用偵測出來後的資訊我們開始做追蹤,因為一些基本背景偵測的問題我們都已假設完整被解決了,例如影子問題或是前景區域破碎的問題都已經解決了,另外我們的系統是以追蹤單人區塊為主,故不去考慮多人追蹤之間的互動,每個人體都是以一個個體獨立去做追蹤的。 第二個部份,人物的身體部位切割。當抽取出移動人物後,為了更加有效進行追蹤人體,我們將人做三區塊的切割,依照人體的比例大致上的切割出我們想要的位置,根據身體部位的不同,因此我們將人物分為三個主要部位。這一部份描述我們為何將人物切成三個身體部位,以及如何分別對此三區塊同時做追蹤。 第三個部份,異常偵測。當我們將人體切割成三區塊之後,我們會分別對此三區塊做區塊的追蹤,那這一部份主要是介紹如何利用這三區塊彼此之間的資訊定義出一自定的限制,以便在測量物體時可以利用此一限制來判斷目前是那塊區塊發生異常。 第四個部份,區塊調整。在第三個部分的步驟結束之後,我們可以得知目標是那一區塊發生異常,因此我們判斷出來要將那一區塊做調整的動作,也就是重新的將異常區塊根據一些簡單的規則以及人體中大致的比例將異常區塊重新拉回相對位置,以及重新做追蹤的動作,這個部分就是討論將區塊調整的規則。 實驗的部份,我們測試了許多影帶並運用我們的追蹤機制。實驗結果發現我們所提出來的機制有著不錯的效果,我們可以利用這個機制來協助來應用在智慧型視訊監控系統(VASM)以及入侵者偵測的運用上。
The purpose of this thesis is to construct a specific person tracking system when given a image with any person. In the surveillance environment, with the progress of the times to process the video signals is more popular .So we think that the VASM is great and satisfies the need of the popular. In tradition we need watch the surveillance system in order to detect the invader non-automatically. So we combine the technology of the image process and the technology of tracking target to support the automatic machine for surveillance automatically In our system, we have four part: detection of moving target in video sequences, decomposing the human body into three block and track each block , abnormal detection, state correction. In the first stage, human detection, we use the foreground detection procedure to detect moving persons in a video. In the foreground detection we assume have solve the shadow problem and while tracking we have know the position of the target we track. In addition, our system is based on the tracking of only one part so we don’t consider the interaction between different blocks, each body is tracked independently. In the second stage, human body decomposition, after detection and separation of the moving suspect in a video sequence, we need to cur the human body into three blocks. Since the weights for matching the body parts may differ, we decompose the human body into three parts. Here, we propose how to cut the human body with three parts. In the third stage, abnormal detection, after cutting the human body into three part we will track the block each so this section to introduce how to define the constraint of the information of three parts. Then when we test some samples, we can detect the abnormal of any block. In the forth stage, state correction, after step three, we will know which part occurs abnormal, so we determine which block need to adjust. In the other words, we use some simply rule to adjust the relative position of abnormal block, and track the new position continuously. In the experiments, we test cases of human tracking , body part segmentation and abnormal detection and state correction. The experimental results show that our system is very effective for human tracking.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT009317599
http://hdl.handle.net/11536/78810
Appears in Collections:Thesis


Files in This Item:

  1. 759901.pdf

If it is a zip file, please download the file and unzip it, then open index.html in a browser to view the full text content.