一個在室內環境下使用環場攝影機與模型比較的多人追蹤系統

自從攝影機發明之後，人們對於監控系統的需求一直存在。我們感興趣的是有效的利用攝影機擷取到的畫面，避免由於太多監控人員長時間待在環境中，所造成之被監控者心裡的不適感，並且能夠一直監視著監控者容易忽略到的死角。也因此，如何能夠在攝影機畫面中，正確的追蹤每一個對象，並加以鎖定監視，便是一個重要的研究議題。　　本系統採用具有360度視角的環場攝影機，加上環場攝影機通常架設點較高(例如天花板)，故和傳統攝影機比較，較不易發生完全遮蔽或是傳統攝影機畫面無法覆蓋的死角，且由於其監視覆蓋面積大，以往需要由2至3台攝影機才能夠完全監控的範圍，只需要由一台環場攝影機便可以達成。本系統所使用之多人辨識方法，避免了舊有方法(例如卡爾曼濾波器法)在環場影像中所可能造成之誤差，採用簡單且迅速的模型比對方式，即時辨識各獨立物件，並持續的保持追蹤，並使用二維高斯濾波器作為空間濾波器，大大的減低由於背景及每個人行走時不同的動作所造成之影響。本系統同時也結合了個人化資訊系統，不但能夠即時的追蹤各個進入環境中之受測者外，更能夠標記各受測者之身分，供使用者參考。　　由實驗結果得知，本論文所提出之系統能夠有效並持續追蹤進入環境中之受測者，並不易由於受測者在環境中交錯行走造成誤判，且可達到9成以上之正確率。如未來系統能結合人臉辨識或Multi-View等方法，或是結合其他可供定位輔助之設備，可以使本論文所介紹之系統定位出受測者更精確之位置並追蹤之。

關鍵字

環場攝影機；多人追蹤；模型比較

並列摘要

The need of surveillance system for human beings always exists after the invention of cameras. We are interested in using the images captured by cameras effectively in order to reduce the uncomfortableness of people being monitored by many spotters on the scene, and keep on monitoring the dead zones that are easily overlooked by the spotters. Thus it is a very important issue to correctly track and lock on each target. The system uses an omni-directional (OD) camera having a 360-degree viewing angle. The OD camera is usually placed at higher place (such as ceiling). Thus, comparing with a traditional camera, an OD camera is less likely to cause a total occlusion, and it also has fewer dead zones. The surveillance coverage of an OD camera is also larger than that of a traditional camera. A surveillance area monitored by 2 or 3 cameras in the past can be done by one OD camera now. The method used in this thesis for identifying multiple people during tracking can avoid the error caused by using traditional methods (such as Kalman Filtering) when OD images are considered. We use a model comparing method which is simple and fast to identify an individual target. We also use a 2-dimesional Gaussian filter to avoid the influence by the background part of the OD image and the different motions associated with each person during walking. The system also combines a personal information system and labels the targets with their entering orders and their personal information. The experimental results show that the system proposed in this thesis can track and identify people effectively and continuously. It is not easy to cause identification errors when people are walking crossover to each other, resulting in more than 90% recognition accuracy. The system can be combined with a face recognition system, a Multi-View method, or other positioning equipments. Then the accuracy of the system can be increased, and the system can locate people more precisely in the surveillance area.

並列關鍵字

Model comparing ； Multiple people tracking ； Omni-directional camera

參考文獻

[2] 宋佩栩，一個使用環場攝影機並結合個人資訊的客製化跌倒偵測系統，中原大學電子工程學系，碩士論文，民國94年。

[3] R. E. Kalman, “A New Approach to Linear Filtering and Prediction problems,” J. Basic Eng., vol. 82-D, pp. 35-45, 1960.

[4] T. Yamane, T. Shiral, and J. Mitrea, “Person Tracking by Integrating Optical Flow and uniform Brightness Regions” in Proc. IEEE Int. Conf. on Robotics and Automation, vol. 4, pp. 3267-3262, May 1998.

[6] H. Roh, S. Kang, and S. W. Lee, “Multiple People Tracking Using an Appearance Model Based on Temporal Color,” in Proc. IEEE Int. Conf. on Pattern Recognition, vol. 4, pp. 643-646, Sept. 2000.

[7] M. Liebens, T. Sakiyama, and J. Miura, “Visual Tracking of Multiple Persons in a Heavy Occluded Space Using Person Model and Joint Probabilistic Data Association,” in Proc. IEEE Int. Conf. on Multisensor Fusion and Integration for Intelligent Systems, pp. 547-552, Sept. 2006.

國際替代計量

一個在室內環境下使用環場攝影機與模型比較的多人追蹤系統

全文下載

主題瀏覽