摘要
在被动定位中,从图像中提取目标的线性旋转不变特征很重要,而非合作目标特征提取要比合作目标困难得多。借鉴圆形目标固有的旋转不变特性,提出了一种虚拟圆方法。该虚拟圆为图像序列相邻帧内三个匹配点经等边三角形向外延拓生成的三个点的外接圆。模拟结果表明,此虚拟圆较其他途径构成的虚拟圆有尖锐的误差分布概率密度函数,进一步研究表明该虚拟圆的直径是一个相当不错的距离相关线段特征。线段特征用于距离估计的特点是简单,虚拟圆概念则增加了它在应用中的灵活性。因为三个匹配点是基于目标特征跟踪的最少匹配点数,故该方法在非合作目标被动定位中具有吸引力。该方法对相邻帧成像时目标相对相机偏斜变化不超过±10°有效。
It is critical to extract the linear rotational invariant of an imaged target in passive ranging from the images, and it′s more difficult to extract the parameters from a general non-cooperative than from a cooperative ones. A method was presented to construct a virtual circle as the inherent rotation invariance of a circular target. The proposed virtual circle was the circumcircle of equilateral by triangles extended from three matched points in adjacent frames in the image sequence. It is demonstrated by the simulations that the probability density function curve of the proposed virtual circle has tighter error distribution than that of a few other methods, and further studies indicate that the diameter of this virtual circle is also a preferable depth-related line segment feature. The line segment feature is used for the target distance estimation and it displays superior performance. It is characterized by its simple for the distance estimation using line segment features, and the concept of virtual circle increases the flexibility in practice. Because as few as three matched points are the least points in target tracking based image feature, so it is attractive for passive ranging to non-cooperative targets. The method is valid under the condition of the inclination angle of target relative to camera increased or decreased from -10 degree to 10 degree between adjacent sampling times.
出处
《红外与激光工程》
EI
CSCD
北大核心
2014年第9期3042-3045,共4页
Infrared and Laser Engineering
基金
国家自然科学基金(60872136,61107007)
陕西省自然科学基金(2011JM8002)
关键词
机器视觉
旋转不变
模式识别
目标跟踪
距离测量
machine vision
rotation-invariant
pattern recognition
target tracking
range finding
