图像稳定算法的性能表征.pdf

PERFORMANCE CHARACTERIZATIONOF IMAGE STABILIZATIONALGORITHMS Stephen B. Balakirsky Rama Chellappa* Army Research Laboratory University of Maryland Adelphi, MD 20783 College Park, MD 20742 ABSTRACT performance in the stabilization of one synthetic infrared This paper considers the characterization of three different data set and two real infrared imagery data sets, along with image stabilization algorithms when used as a preprocessor a stable infrared baseline data set. All of the real sequences for a computer vision application. These algorithms vary in used for evaluation were 100frames long, taken at a frame computational complexity, accuracy, and performance. The rate of ten frames per second. This paper concentrates on first algorithm (developed by the Amy Research the performance characterization with respect to computer Laboratory (ARL)) is capable of image alignment to an observed imagery, where computer-observed imagery is accuracy of one pixel. Algorithms 2 and 3 (developed by defined as a video stream that is used as input into a the University of Maryland) are capable of full subpixel computer vision application. Thus the computer is stabilization with respect to translation, rotation, and scale. “observing” the scene using the image stream as its The evaluation tools incorporated include mean square observations. ARL, in cooperation with the Advanced error of the output data set and the overall