基于角点配准的多分辨率背景估计与补偿算法

Multi-resolution background estimated and compensation based on corner point matching

本文针对动态背景下序列影像的目标检测,提出了基于改进NCC的Harris角点配准和基于多分辨率的全局运动参数估计方法:通过对相关度量NCC加入距离度量信息,寻求相邻图像帧中的正确匹配点对,再根据构造的分辨率影像,估计全局运动参数,以此进行运动补偿,从而利用差分法可准确地检测出运动目标。实验结果表明,该方法能够有效地实现背景估计和全局运动补偿,自动检测出视频序列中的运动目标。

This paper studied on moving target detection with non-stationary background. The method of registration based on Harris with improved NCC and global motion parameter estimation based on multi- resolution was proposed. It searched for the correct pairs of points in adjacent image by adding distance measurement information to NCC which is correlativity information. Then it estimated the global motion parameter estimation with multi-resolution image, and motion compensation was used. With all above, moving target could be detected accurately based on frame-differencing. Experimental results demonstrated that this method could effectively get background estimated and compensated, and extract moving target in the video sequence.