透過您的圖書館登入
IP:3.91.8.23
  • 學位論文

以環場與PTZ攝影機進行移動物件即時追蹤之監控系統

Using Omni-Directional and PTZ Cameras to Implement a Surveillance System for Real-Time Tracking of Moving Objects

指導教授 : 繆紹綱

摘要


摘要 隨著數位影像處理技術與DSP處理器的發展,智慧型視訊監控系統的探討越來越受重視。以往用作監控的攝影機在一個時間點上可監控到之範圍過於狹小,或有攝影機無法照到之死角(如攝影機正下方),因此本論文採用同一時間可環視全場(omni-directional)之監視系統。當有目標物進入時,此系統在第一時間能即時偵測出,再搭配PTZ攝影機加以追蹤,鎖定目標物並輸出其影像供查看。 本論文所提監控系統的特色在於結合了環場與PTZ兩種功能截然不同的攝影機,且未使用到複雜的演算法,大多只運用基本的影像處理方法來達到即時監控的目的。換言之,本論文利用互補性強的兩種攝影機,搭配簡單的偵測與追蹤演算法。在偵測追蹤的過程當中,本系統先從環場攝影機所擷取到的連續環場影像中框選目標物及其所在座標,之後將此座標轉換到PTZ攝影機的座標,並持續控制PTZ攝影機指向目標物並擷取其適當大小的影像。 本系統先於PC上模擬,再由DSP發展板實現。在座標轉換上,x與y方向上的平均誤差均約為6個pixels,換算回空間中實際距離則約為6 cm左右(影像大小為752×771,實際空間大小亦為752×771 cm2),這相當於分別是0.80%與0.78%的誤差。在鎖定目標物要佔整個影像適當比例的實驗中,未有採用與有採用自動縮放功能演算法之比例的變異數估測分別為22.32及4.24,由此得知使用自動縮放功能可達成固定人物在影像中出現比例的目的。此外由於PTZ攝影機在機械結構上轉速的限制而不能夠即時的在時間內有效的傳送指令驅動攝影機。實驗結果顯示大約每6張畫面傳送一次指令給PTZ攝影機是最有效率的傳送頻率,在此時間期間可以進行座標轉換與目標物追蹤,而這也限制了可被成功追蹤之目標物的移動速度;在距離PTZ攝影機五公尺處,可允許速度為130 cm/sec,到2公尺時降為50 cm/sec。

並列摘要


Abstract With the development in digital image processing technology and DSP processors, the discussion on intelligent video-information surveillance systems receives more and more attention. The viewing angle of a traditional surveillance camera may be too narrow at a time instant and it may also have the dead-zone problem. Thus, this thesis develops a surveillance system based on the use of an omni-directional (OD) camera that can look around the whole surveillance area at the same time. The system can detect a target as soon as it enters the surveillance area and then track and lock it based on the use of a Pan/Tilt/Zoom (PTZ) camera. The proposed system in this thesis is characterized by combining the use of an OD camera and a PTZ camera and by using only basic image processing methods instead of complex algorithms. In other words, we use two complementary cameras along with simple object detecting and tracking methods. During the object tracking process, the system gets the coordinates of the moving object in consecutive frames captured by the OD camera. Then the system transforms the coordinates for the OD camera to that for the PTZ camera and guides the PTZ camera to point to the object with proper pan, tilt, and zoom parameters. We simulate the proposed system in PC, and then implement it on a DSP board. For coordinate conversion, the average errors in x-direction and y-direction are both about 6 pixels, which correspond to 6 cm in real space (the image size is 752×771). So the errors are about 0.80% and 0.78% in x- and y- directions, respectively. In the experiment which controls the ratio of a person’s size appearing in an image, the resulting variances are 22.32 and 4.24 for not using and using the proposed zoom function, respectively. The results show that the zoom function can achieve the objective of fixing the ratio of person’s size appearing in the image. We found that due to nonzero response and execution time of a PTZ camera, it can not accept and execute consecutive commands effectively. The experimental results show that the most effective frequency is to send only one command in 6-frame time to the PTZ camera so that we can do the coordinate transform and the target tracking within this time duration. This in turn limits the allowed moving speed of the target that can be tracked successfully; for 5 meters away from the camera, the allowed speed is 130 cm/s, and it drops to 50 cm/s from 2 meters away.

參考文獻


[1] G. D. Hong, The Study of Integrating RFID Technology with Image Analysis in the Elders’ Falling Detection, Master Thesis, Department of Information Management, Fu Jen Catholic University, Taiwan, 2006.
[35] T. S. Huang, A Smart Digital Surveillance System with Face Tracking and Recognition Capability, Master Thesis, Dept. of Electronic Eng., Chung Yuan Christian University, 2004.
[2] M. Greiffenhagen, D. Comaniciu, H. Neimann, and V. Ramesh, “ Design, Analysis, and Engineering of Video Monitoring Systems: An Approach and A Case Study,” Proc. of IEEE, vol. 89, no. 10, pp. 1498-1517, Oct. 2001.
[3] S. Morita, K. Yamazawa, and N. Yokoya, “Networked Video Surveillance Using Multiple Omnidirectional Camera,” Proc. of IEEE Int. Symposium on Computational Intelligence in Robotics and Automation, pp. 1245-1250, Kobe, Japan, 2003.
[5] P. K. Allen, B. Yoshimi, and A. Timcenko, “Real-Time Visual Servoing,” Proc. of IEEE Int. Conf. on Robotics and Automation, pp. 851-856, April 1991.

延伸閱讀