結合顏色與空間特性的物件追蹤演算法

Object Tracking Combining Chromatic and Spatial Features

孫維辰(James)

指導教授 ： 楊士萱

國立臺北科技大學/電資學院/資訊工程系研究所/碩士(2009年)

https://doi.org/10.6841/NTUT.2009.00350

摘要

物件追蹤的目的主要是在追蹤影像序列中的視覺物件，其結果已廣泛使用於物件辨識、自動監控、人機互動、視訊索引、以及視訊傳輸等應用中。物件追蹤常用的物件特徵有顏色、邊緣、光流、紋理、與特徵點等，傳統的物件追蹤演算法（如CamShift），在背景與追蹤物件顏色相近，或追蹤物件有較複雜的顏色、光影、或形狀變化時，通常無法提供可靠的追蹤結果。本論文整合數種物件特徵與物件追蹤技術，包括調整色彩空間和抑制背景顏色之改良CamShift演算法，配合物件中心加權，以及區域成長演算法的輔助，降低追蹤錯誤的可能性。我們利用precision和recall來評估物件追蹤的正確率， CamShift演算法的平均正確率約為40%，而我們所提出之改良演算法的平均正確率則達80%以上。

關鍵字

物件追蹤 ； 色彩空間 ； CamShift ； SIFT

並列摘要

Object tracking is the process of location moving objects from image sequences, which has found wide applications in object recognition, automatic surveillance, human-computer interface, visual indexing, and video coding. Conventional object-tracking techniques do not always provide reliable tracking results when the tracked object has very similar features(such as colors) to the background. To overcome this problem, in this paper we propose a series of background-suppression techniques to be integrated with the conventional CamShift algorithm. The regions in the foreground with the same colors as the background will be suppressed in the first place, and then identified and compensated by centroid weighting and region growing. We also propose using an improved color quantization scheme to increase the color discrepancy. The accuracy of a tracking algorithm is measured by the associated precision and recall rates on a frame-by-frame basis. The proposed algorithm achieves averagely more than 80% accuracy for the examined test sequences, which is significantly better than the conventional tracking algorithm based on CamShift.

並列關鍵字

object tracking ； color space ； CamShift ； SIFT

參考文獻

[1] Y. Alper, J. Omar, and S. Mubarak, “Object tracking: a survey,” ACM Computing Surveys, vol. 38, 2006.

[3] J. F. Canny, “A computational approach to edge detection”, IEEE Trans. on PAMI, pp. 679–698, 1986.

[7] K. I. Laws, “Textured Image Segmentation. Ph.D. thesis,” University of Southern California, 1980.

[8] S. C. Zhu and A.Yuille, “Region competition: Unifying snakes, region growing, and Bayes/MDL for multiband image segmentation,”IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 18, no. 9, pp. 884–900, Sep. 1996.

[11] D. G. Lowe, “Distinctive Image Features from Scale-Invariant Keypoint,” International Journal of Computer Vision, Vol. 60, No. 2, pp. 91-110, 2004.