基于扩展卡尔曼滤波的空间小目标跟踪算法

An Algorithm for Small Space Target Tracking Based on Extended Kalman Filter

由于天基平台拍摄天空图片时,背景和相机同时发生相对运动,造成相邻帧之间无法通过简单的帧差法得到运动的小目标,造成了空间目标检测的难度。在分析星空图像模型的基础上,提出了一种提取特征点组成匹配三角形的图像配准方法,该方法通过最优阈值的选取对单帧图像进行分割,去除背景噪声。将星点按面积大小划分,符合条件的星点组成特征三角形并在相邻帧中进行匹配得到运动参数。在配准时为了减小计算量,忽略背景插值只针对星点坐标矩阵进行处理。最后通过多帧轨迹关联检测出目标的运动轨迹。仿真实验表明,在运动的序列图像中,该方法能实现高检测率和低虚警率的实时检测。

When the space-based platform shoots the star-image pictures,the background and the camera have relative motion simultaneously. Thus the small moving targets can not be detected through the simple frame difference. Based on analysis to star image model,an image registration method is proposed by extracting feature points to match the same triangle. In the method,optimum threshold is selected for single frame image segmentation and background noise removal. Then,the bright stars are divided by the area,and the feature triangles are made up by the stars that meet the conditions. The motion parameters can thus be obtained in adjacent frames by matching the feature triangle. For reducing the complexity,the star centroid coordinate matrix is processed instead of the whole star image to image registration. Finally,the target track is obtained through a multi-frame track association. Simulation and tests demonstrate that the method can achieve high detection rates and low false alarm rate in the sequence frames.