复杂背景下的快速机动目标跟踪算法  

Fast-moving maneuvering target tracking algorithm under complicated background

作　　者：谢泽奇[1] 张会敏[1] 张佳佳 张云龙[1] 张善文[1] XIE Ze-qi;ZHANG Hui-min;ZHANG Jia-jia;ZHANG Yun-long;ZHANG Shan-wen(Sias International University of Zhengzhou University, Zhengzhou 451150, China;Laboratory of Parallel Software and Computer Science, Institute of Software, Chinese Academy of Sciences ,Beijing 100190, China)

机构地区：[1]郑州大学西亚斯国际学院,河南郑州451150 [2]中国科学院软件研究所并行软件与计算科学实验室,北京100190

出　　处：《传感器与微系统》2018年第5期132-134,143,共4页Transducer and Microsystem Technologies

基　　金：国家自然科学基金资助项目(61272333);河南省科技厅科技攻关研究项目(182102210545);河南省教育厅科学技术重点研究项目(16A520095)

摘　　要：为了解决复杂背景及大视野场景下跟踪机动目标易丢失和跟踪精度低的难题,提出了一种复杂背景下的快速机动目标检测与跟踪算法。利用帧间差分算法提取图像中的机动目标,在初始帧建立机动目标的颜色直方图模型,将后续输入图像的像素值转化为直方图分布下的概率值;根据与目标模型的相似度,将每个候选区域的像素值作为密度;利用自适应均值漂移算法寻找机动目标的真实位置;利用卡尔曼滤波预测目标位置。实验结果表明:算法能够准确地在复杂背景和大视野场景下快速检测并跟踪机动目标。In order to solve the problem that tracking maneuvering target under complex background and large view is easy to lose and the problem of low tracking precision, propose a fast maneuvering target detection and tracking algorithm under complicated background. Frame difference algorithm is utilized to extract maneuvering target in image, and color histogram model of maneuvering target is established in initial frame, and the subsequent input image pixel values transformed into probability value under histogram distribution, according to the similarity with the target model and pixel values of each candidate region as density. Using adaptive mean shift algorithm to search for search real position of maneuvering target. In the process of tracking. Using Kalman filtering to predict target position. The experimental results show that the algorithm can accurately fastly detect and track maneuvering targets under complex background and large view.

关 键 词：全景图像 快速跟踪 颜色直方图 均值漂移 卡尔曼滤波 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]