摘要
多目标跟踪往往是在复杂的自然背景当中检测若干个弱小机动目标。根据自然背景和弱目标的频率特性,提出了一种基于小波变换的小波能量检测目标的方法。目标在水平、垂直两个方向上的小波能量要远高于背景。再经过自适应域值选取,将目标和背景分割开来。用分类法计算每个目标的形心,根据相邻几帧各目标形心,可用扩展卡尔曼滤波器来估算下一时刻各目标的位置和速度。经仿真实验证明,该方法能够检测出海、空等自然背景中多个弱小目标,扩展卡尔曼滤波权值á和a取为0.8时,目标横、纵坐标误差小于1个像素,速度误差小于1个像素/帧。
Multi-target tracking is usually to detect faint targets in complex background. A method for detecting targets with wavelet energy is proposed based on frequency characteristics of natural background and faint target. The wavelet energy of a target both in horizontal and vertical directions is far higher than that on background. The targets can be separated from background through adaptively selecting regional-value. The centroid of each target can be calculated by means of classification. The positions and velocities of various targets at next frame can be estimated by extended Kalman filter according to centroids of various targets in adjacent several frames. Simulation experiments demonstrate that multiple faint targets can be detected from ocean and sky natural background by this method. When both α and β of the extended Kalman filter is 0.8, errors of transversal and longitudinal coordinates are less than 1 pixel and the velocity error is less than 1 pixel/frame.
出处
《光电工程》
CAS
CSCD
北大核心
2004年第11期16-19,共4页
Opto-Electronic Engineering
基金
中国科学院长春光学精密机械与物理研究所创新基金(ZJ99I30B)
关键词
多目标跟踪
小波变换
目标检测
卡尔曼滤波
Multi-target tracking
Wavelet transform
Target detection
Kalman filtering