基於運動視訊中動態背景之移動物件偵測與追蹤研究

Research on Moving Object Detection and Tracking in Dynamic Background Using Sport Videos

孫琬婷(Wan-Ting Sun)

指導教授 ： 張元翔

中原大學/工學院/資訊工程研究所/碩士(2013年)

https://doi.org/10.6840/cycu201300645

摘要

由於多媒體產業的蓬勃發展，運動視訊的取得已不再僅限於電視轉播，視訊分享網站也提供許多運動視訊。運動視訊中偵測及追蹤移動物件，是相當具有挑戰性的問題。本研究提出一套「基於運動視訊中動態背景之移動物件偵測與追蹤」系統，系統流程主要包含：(1) 特徵點擷取；(2) 移動向量估測；(3) 移動向量分群；(4) 移動物件偵測；及 (5) 移動物件追蹤。研究方法主要是採用金字塔Lucas-Kanade 技術，結合K-Means分群法，將特徵點進行分群，進而偵測移動物件，並利用卡爾曼濾波器進一步預測及修正偵測結果，強化移動物件追蹤的準確率。研究結果顯示，在動態背景下之運動視訊中，系統的平均偵測率可達83%，透過卡爾曼濾波器修正後之平均追蹤準確率可達到90%。總結而言，本研究提出的方法可用於運動視訊中動態背景之移動物件偵測與追蹤，進一步分析運動視訊中運動選手的運動軌跡及行為，未來的應用層面也相當廣泛。

關鍵字

電腦視覺 ； 移動估測 ； 分群演算法 ； 視訊追蹤

並列摘要

As the rapid development of multimedia industry, acquisition of sport videos is no longer limited to TV broadcast. Video-sharing websites also provide many sport videos. Detection and tracking of moving objects in sport videos is a challenging problem. This study proposed a system for moving object detection and tracking in dynamic background using sport videos. The system processes mainly include: (1) feature point extraction ; (2) motion vector estimation; (3) motion vector clustering; (4) moving object detection; and (5) moving object tracking. The research method is to use the pyramid Lucas-Kanade technology, combined with the K-Means clustering for feature points. The Kalman filter is used for further prediction and correction of detection results, thereby improve tracking accuracy for moving objects. The results show that our system could achieve average detection rates of 83%, and could further achieve average tracking accuracy of 90% in sport videos with dynamic background. In conclusion, our proposed method could be used for detection and tracking of moving objects, and further be used for the analysis of the moving trajectories and behavior of athletes. Future application aspects are also very broad.

並列關鍵字

video tracking ； motion estimation ； lustering ； computer vision

參考文獻

[1] A. Bugeau, and P. Pérez, “Detection and segmentation of　moving objects in highly dynamic scenes,” Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, 2007.

[2] D. Szolgay et al., “Detection of moving foreground objects in videos with strong camera motion,” Pattern Analysis and Applications, Vol. 14, No.3, pp.311-328, 2011.

[3] J. Zhong and S. Sclaroff, “Segmenting foreground objects from a dynamic textured background via a robust Kalman filter,” Proceedings of Ninth IEEE International Conference on Computer Vision, Vol. 2, 2003.

[4] M. Sankari, and C. Meena, “Estimation of dynamic background and object detection in noisy visual surveillance,” International Journal of Advanced Computer Science and Applications, Vol. 2, No. 6, 2011.

[6] A. Talukder, et al., “Real-time detection of moving objects in a dynamic scene from moving robotic vehicles,” Proceedings of on IEEE International Conference on Intelligent Robots and Systems (IROS), Vol. 2, pp. 1308-1313, 2003.