基于小波变换与卡尔曼滤波的多目标跟踪  

Multi-object Tracking Based on DWT and Kalman Filter

作　　者：赵红怡[1] 韩燕[1] 

机构地区：[1]北方工业大学信息工程学院,北京100144

出　　处：《北方工业大学学报》2014年第3期11-15,77,共6页Journal of North China University of Technology

摘　　要：利用小波变换和卡尔曼滤波研究多目标检测与跟踪问题.采用小波三层分解原理,对图像进行去噪处理,得到的低频图像再进行背景差分运算,从而检测出运动目标.采用卡尔曼滤波先预测出目标在下一帧的位置,通过前后帧的目标位置计算半径值,结合k近邻算法在半径内采用k近邻数据关联,取与预测位置欧氏距离最短的点为目标在下一帧中出现的真实位置.通过MATLAB对实验进行仿真.结果表明,方法可有效提高多目标检测与跟踪的准确性和抗遮挡性.DWT（Discrete Wavelet Transform） and Kalman filter is used to research the detec tion and tracking of the multi object. The three layer wavelet decomposition principle is employed to make image denoising processing, and then use the background difference method to detect the multi-object from these low-frequency images. By applying the Kalman filter, the positions of ob- jects in the next frame are predicted and the radius value is calculated according to the target posi- tion of the previous and following frames. Then, a shortest Euclidean distance point is found out which is the true position of the next frame target by KNN （k-Nearest Neighbor algorithm） data as- sociation. The MATLAB simulation comprises that the method can effectively improve the anti- blocking and accuracy of multi-object detection and tracking.

关 键 词：多目标跟踪 离散小波变换 卡尔曼滤波 K近邻 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]