基于光流计算模型的智能电子稳像的研究.pdfVIP

第 32 卷 第 12 期增刊 仪 器 仪 表 学 报 Vol.32 No.12 2011 年 12 月 Chinese Journal of Scientific Instrument Dec. 2011 * 基于光流计算模型的智能电子稳像研究 崔玉柱 王 睿 李敬祯 （北京航空航天大学仪器科学与光电工程学院 精密光机电一体化教育部重点实验室 北京 100191） 摘 要：介绍了光流计算模型下智能电子稳像算法的整个流程，包括三部分：运动估计、抖动检测和运动补偿。重点介绍了 运动估计算法和智能抖动检测算法；详细阐述了在OpenCV平台上，首先对图像序列进行高斯滤波，然后利用金字塔Lucas Kanade 算法进行全局运动估计并设置判别条件，进行智能抖动检测，最后对发生抖动的序列进行图像补偿，从而得到稳定 的视频序列的电子稳像方法。利用结构相似度—SSIM 对稳像的结果进行了客观评价。实验数据及结果的分析表明：平移全 局运动估计精度能达到亚像素，旋转估计精度能达到0.5°以下；稳像后计算得到的结构相似度SSIM值平均能达到0.96左 右，说明我们采用的稳像策略效果良好。 关键字：电子稳像 金字塔LucasKanade光流法 运动估计 结构相似评价 Intelligence Video Stabilization Based on Pyramid Optical Flow Algorithm Cui Yuzhu Wang Rui Li Jingzhen (Laboratory of Precision Opto-mechatronics Technology, Ministry of Education, School of Instrumentation and Optoelectronic Engineering, Beihang University, Beijing 100191) Abstract ：The method of electronic image stabilization, which includes motion estimation, jitter detection and motion compensation is introduced. Not only the motion estimation and jitter detection are focused on, but also the main process of the method which include: Gaussian filtering the image sequences and the global motion estimation vector with pyramid Lucas Kanade algorithm, how to set thresholds to detect if unwanted vibrations existed, as well as the image compensation strategy used to translate and rotate the shaky image to the right place to achieve a stabilized video with OpenCV is introduced. The experimental results show that the accuracy of motion estimation of translation and rotation can reach sub-pixel and 0.5o respectively. SSIM results of image sequences average 0.96 which shows that the proposed electronic stabiliza