Abstract
In this paper, we study the problem of subsequence matching and address the efficiency issue in large-scale video retrieval. The subsequence matching task can be transformed to a 2D Hough space projection of pairwise frame similarities between two subsequences. We present an approximate Hough transform that replaces the 2D Hough space with a 1D Hough space, and thus reduce the time complexity for voting and searching over the Hough space. We apply the methodology in video stream monitoring where every stream subsequence is matched with a large amount of the reference dataset. The experiment compares with some state-of-the-art methods and the result shows the superior of our method in terms of efficiency.