仿射不变特征转换的运动目标跟踪方法  

Motion object tracking based on Affine SIFT

作　　者：刘景正[1] 余旭初[1] 李艳 

机构地区：[1]信息工程大学测绘学院 [2]63961部队

出　　处：《测绘科学》2012年第4期26-28,共3页Science of Surveying and Mapping

基　　金：国家自然科学基金项目(编号40971239)

摘　　要：运动目标跟踪技术在交通管制、智能监控、GPS导航等多个领域有广阔的应用前景。本文提出一种基于经纬度和尺度的ASIFT(Affine-SIFT)跟踪算法,模拟序列图像成像时的倾斜角度,进行实时更新来去除由形变带来的过时特征信息,保证了特征的稳定,提高了配准精度。实验证明,此方法不受传感平台倾斜造成的图像仿射变形影响,能够有效处理目标由于旋转、形变而导致跟踪性能下降甚至跟踪目标丢失的问题,同时对跟踪过程中多目标的关联,也具有较好的鲁棒性,取得了良好效果。The moving object tracking technology has a broad application prospect in the domain of traffic control,intelligent control,GPS navigation and other fields.The moving objects on sequential imagery acquired by unmanned aerial vehicle with large rotation and tilt deformation,due to instability of flying,make it difficult to track.ASIFT（Affine-SIFT） tracking algorithm based on latitude,longitude and scale was introduced in this paper,in which the imaging tilt angle was simulated,and different affine images were normalized to a unification model,maintaining affine invariant.Experimental results showed that this method could resist the influence of image affine distortion caused by the tilt sensor flat,and not only handle the problems of object losing effectively,which are induced by object rotation and translation,but also have nice robustness to the conjunction of multi-objects in the process.

关 键 词：序列图像 ASIFT 图像匹配 运动目标 目标跟踪 

分 类 号：P235.2[天文地球—摄影测量与遥感]