基于图像融合的动态轮廓线跟踪新方法  

A New Tracking Method of Dynamic Contour Based on Image Fusion

作　　者：赵鹏[1] 浦昭邦[1] 张田文[2] 

机构地区：[1]哈尔滨工业大学自动化测量与控制系,哈尔滨150001 [2]哈尔滨工业大学计算机科学与工程系,哈尔滨150001

出　　处：《光学学报》2005年第6期760-766,共7页Acta Optica Sinica

基　　金：国家自然科学基金(69775007;60075010)资助课题。

摘　　要：红外与可见光传感器是目标跟踪识别系统中常用的两种传感器,对这两种传感器图像进行融合能有效提高系统跟踪检测的准确性。将动态轮廓线模型与图像融合结合,在特征搜索过程中利用特征点准确地完成了图像配准,同时使用了一种新的特征级融合方法,将两种图像中目标轮廓的B样条曲线控制点进行实时微分耦合。这种耦合将Curwen提出的微分耦合机制作了改进,利用图像配准把刚性硬模板改变为实时的变换模板并推导了融合后动态轮廓线的新的动力学方程。这种融合利用了红外图像目标轮廓信息约束可见光图像中动态轮廓线的收敛形状,有效地提高了可见光图像目标跟踪的准确性。对运动人手序列图像的对比跟踪实验表明,这种融合使得可见光图像中动态轮廓线平均跟踪误差减小了60.25%。Infrared (IR) and visible sensors are commonly used in the target tracking and recognition system. Image fusion for these two modal images can effectively improve the system's tracking and detection accuracy. The model of dynamic contour is combined with image fusion and feature points in feature search are used to implement image registration accurately. Meanwhile a new feature-level image fusion is applied. Control points of B-spline curves for the target's contour in two modal images are used to implement a real-time differential coupling. This coupling makes improvement on the differential coupling proposed by Curwen, where a rigid template is transformed into a real-time transformation template with image registration. Moreover, a new dynamic equation is derved for dynamic contour after this image fusion. In this fusion, the dynamic contour's convergent shape in visible image is restricted by the target's contour in IR image. This fusion improves dynamic contour's tracking accuracy effectively. A contrasting experiment on moving hand image sequence indicates average tracking error of dynamic contour has decreased by 60.25% in visible image with this image fusion.

关 键 词：信息光学 图像融合 视觉跟踪 动态轮廓线 微分耦合 

分 类 号：TN911.73[电子电信—通信与信息系统]