利用多台俯視深度相機進行室內人員識別與追蹤之監視系統

Person Identification with Tracking System for Indoor Surveillance Using Multiple Top-view Depth Cameras

蕭伯豪(Po-Hao Hsiao)

指導教授 ： 傅立成

國立臺灣大學/電機資訊學院/電機工程學研究所/碩士(2014年)

https://doi.org/10.6342/NTU.2014.02557

摘要

在這篇論文中，我們提出了一套新的人類識別方式，希望透過在天
花板上所架設的俯視深度相機所得到的生物特徵可以完成室內人員的
識別。在最一開始從深度相機得到結果我們會在前置作業中先透過背
景相減，來得到不屬於背景的部分，接著我們會透過座標軸的轉換修
正相機中物體的形變。透過人形的偵測我們會先將在前景中不屬於人
類的部分去除，針對偵測的結果我們會從深度影像中所得到的生物特
徵所得到的特徵向量透過支持向量機的方式去進行室內的人員識別，
對於識別的結果透過粒子濾波器去追蹤人形。我們所提出的架構有以
下三項優點: 1) 偵測器可以即時挑選出正確的人形使識別更加有效率。
2) 可以對於影像中出現的多個人形進行同時識別 3) 透過追蹤器的加
入，我們可以抵抗高度變化或劇烈形變。
在結果中我們展示了系統的準確程度，我們就人員辨識成功率，透
過混淆矩陣的方式作為實驗的呈現，驗證了整個系統的有效性。

關鍵字

人員識別 ； 生物特徵 ； 粒子濾波器

並列摘要

In this thesis, we propose a novel person identification system, hoping to
identify members at indoor environment by using the biological feature
obtained by top-view depth camera. In the beginning, we do the background
subtraction from the original depth information image and get the
foreground object. After extracting the foreground image we will use a
simple mapping function from camera coordinate to the world coordinate. A
top-view human detector is used to understand which foreground object is
really human. The SVM classifier will be applied to identify people after the
detection process. In the end, the SIR particle filter will be utilized to track
the human. There are three advantages of our framework: 1) the detector
can pick up true human shape in real-time which can make our identification
process more effective, 2) our proposed person identification system can
simultaneously identify multiple people in the same time, 3) according to the
addition of the tracker, we can tolerate height variation and human shape
distortion. At last, in the experimental results, we use the confusion matrix
to evaluate effectiveness of our person identification process, and
successfully validates the accuracy of the system.

並列關鍵字

Human identification ； biological feature ； particle filter

參考文獻

people detection and tracking for indoor surveillance using multiple top-view depth

pp. 1739–1742, Nov 2012.

probabilistic person identification in tv-series,” in IEEE Conference on Computer

Vision and Pattern Recognition (CVPR), pp. 2658–2665, 2012.

[4] A. Jain, L. Hong, and S. Pankanti, “Biometric identification,” Commun. ACM,