基于角点跟踪的数字稳像算法  

Digital image stabilization algorithm based on corner tracking

作　　者：尹德森[1] 赵跃进[1] 

机构地区：[1]北京理工大学信息科学技术学院光电工程系,北京100081

出　　处：《红外与激光工程》2008年第5期920-923,共4页Infrared and Laser Engineering

基　　金：国家部委基金资助项目(03001070302)

摘　　要：在基于特征跟踪的数字稳像算法中,特征提取不稳定和特征匹配计算量大是目前存在的主要问题。针对这一问题,利用角点检测稳定性较好的Harris算子,在图像局部区域内提取角点,通过对比在当前图像中定义的角点距离与在参考图像中定义的窗口距离来确定角点间的对应关系,进而计算仿射模型的运动参数,最终进行运动补偿以达到输出稳定视频图像的目的。实验结果表明:该方法具有很好的稳定性和准确性,并有效地减少了算法的运行时间。因此,这种稳像算法适合于实际场景中的实时应用。Feature extraction instability and feature matching complexity are two main problems in digital image stabilization methods based on feature tracking. A novel digital image stabilization method, which used Harris operator with good stability in comer detection, was proposed. This method extracted the comer features in local image region firstly. Then, by comparing the comer distance in current image and the window distance in reference image, the corresponding relationship between comers were confirmed. Next, the motion parameters of affine model were computed using these corresponding comers. Finally, stable video output were obtained by motion compensation. The experimental results show that the proposed mehod can obtain good accuracy, stability and computational time efficiency. As a result, this method is suitable in real image stabilization tasks.

关 键 词：数字稳像 HARRIS角点检测 运动估计 运动补偿 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]