空拍視訊之人群監測的演算法及應用

Algorithms and applications on medium-scale human crowd surveillance in aerial videos

邱漢坤(Han-Kuen Chiu)

指導教授 ： 王家祥

國立清華大學/電機資訊學院/資訊工程學系/碩士(2012年)

https://doi.org/10.6843/NTHU.2012.00521

摘要

無人駕駛機上的攝影機所建置之視覺輔助系統已廣泛被作為各種應用，例如用於城市、廣場或交通的監控，然而目前在空中平台卻沒有一套有效的人體辨識系統。其原因是機體或環境的不穩定導致影像的某些人物會暫時消失，還有從高空拍攝的人物會變得極小。而現有追蹤技術都專注於高階特徵的擷取，例如人臉清晰的五官或人體動作的辨識，因此不適用於空拍視訊。本研究旨在於解決此困難，我們提出了演算法內容包含利用動作的預估以及空間的相關性做猜測，並結合權重式區域匹配做人體辨識，來達到監控人群的效果。除此之外，我們將已辨識的人群逐一加入資料庫中，並附加一些資訊及注釋，讓系統有更多其他的應用，並得到群體或個人的一些統計量。經由實驗可以得知，我們提出的人形辨識方法在低解析度的人形影像仍可達到90%以上的準確率，並且與權重式區域匹配的方法相比更能達到即時辨識。

關鍵字

人體辨識 ； 空拍視訊 ； 監測 ； 權重式區域匹配

並列摘要

The greatest use of vision system on UAVs has been in the areas of surveillance purposes such as civil applications. However, there is still not a robust human identification method in aerial platform because highly vibration causes unstable videos and persons temporarily out of field. Another reason is that medium altitude videos contain low resolution persons, a great deal of moving persons. Previous multi persons tracking or identification methods have often considered by key characteristics recognition such as face or human pose, specific high level invariant features. The goal of this research is to solve these problems. The algorithm would combine the temporal and spatial cue in videos with weighted region matching (WRM) method. Besides, we maintain a comprehensive exemplar database that stores the retrieved human regions with rich human annotations, which make our system utilize in more civil applications and get some statistics of individuals and groups. According to the experimental results, the proposed human blob identification method is effective in spite of the low resolution human figures. In the meanwhile, average time consuming in matching process is less than origin WRM method and enhance real-time.

並列關鍵字

person re-identification ； aerial video ； surveillance ； weighted region matching

參考文獻

[1] C. S. Sharp, O. Shakernia, and S. S. Sastry, " A Vision System for Landing an Unmanned Aerial Vehicle," IEEE International Conference on Robotics, pp. 1720-1727,2001.

[2] P.-J. Bristeaui, F. Callou, D. Vissière, and N. Petit, "The Navigation and Control technology inside the AR.Drone micro UAV," IFAC World Congress Milano, pp. 1477-1484, 2011.

[3] I. Cohen and G Médioni, "Detection and tracking of objects in airborne video imagery," Technical report, University of Southern California, 1998.

[4] Z. Yue, D. Guarino, and R. Chellappa, "Moving object verification in airborne video sequences," IEEE Transaction on Circuits and Systems for Video Technology, vol. 19, pp. 77-89, 2009.

[5] Y. Huang and X. Luo, "Simultaneous Detection and Tracking in Airborne Video," International conference on computer technology and development, pp. 320-324, 2009.