首页 | 本学科首页   官方微博 | 高级检索  
     

采用Kalman_BP神经网络的视频序列多目标检测与跟踪
引用本文:曲仕茹,杨红红. 采用Kalman_BP神经网络的视频序列多目标检测与跟踪[J]. 红外与激光工程, 2013, 42(9): 2553-2560
作者姓名:曲仕茹  杨红红
作者单位:1.西北工业大学 自动化学院,陕西 西安 710129
基金项目:航天科技创新基金(CASC201104);航空科学基金
摘    要:针对在复杂环境下多目标检测与跟踪实时性差和准确率低的问题,提出了一种基于神经网络修正均方误差估计的卡尔曼滤波跟踪方法,实现视频序列的多目标跟踪。在该方法中,首先通过帧间差分法准确提取出背景,并结合背景消减法实现多目标的检测,应用形态学滤波对检测结果进行优化;然后利用Kalman_BP神经网络预测滤波器对运动目标的位置进行预测。BP神经网络的引入,主要是降低由于模型变化以及噪声等引起的Kalman滤波器的估计误差,使Kalman滤波器的预测结果更加精准;最后,通过对不同的目标贴上标签,实现目标快速匹配,根据相邻帧间同一目标形心位置以及外接矩形的一致性,建立目标链,实现多目标跟踪。实验结果表明,该算法不仅能够快速稳定地对不同场景中的目标进行跟踪,而且能够统计目标数目和显示目标的运动轨迹,与粒子滤波等方法相比跟踪更加平稳,提高了跟踪的可靠性。

关 键 词:多目标检测   多目标跟踪   Kalman滤波   BP神经网络
收稿时间:2013-01-11

Multi-target detection and tracking of video sequence based on Kalman_BP neural network
Qu Shiru , Yang Honghong. Multi-target detection and tracking of video sequence based on Kalman_BP neural network[J]. Infrared and Laser Engineering, 2013, 42(9): 2553-2560
Authors:Qu Shiru    Yang Honghong
Affiliation:1.School of Automation,Northwestern Polytechnical University,Xi'an 710129,China
Abstract:To improve the recognition rate and speed of the multi-target detection and tracking in the complex background, a tracking method based on neural network Kalman filter with correction mean square error estimation was proposed. Multi-target detection and tracking of the video sequence were achieved. In this method, first of all, the background was extracted accurately through the inter-frame difference method and multi-target detection was achieved combined with background subtraction method,the detection results were optimized utilizing morphological filtering. Then, Kalman_BP neural network filter was used to predict the position of the moving target. The estimation error of the Kalman filter caused by model changing and noise was mainly reduced with BP neural network, which made the predictive results more accurate. Finally, the fast matching of target was achievid via labeling different targets. Target chain was established by using the characteristics that little change of same goal centroid position and the boundary rectangle between the adjacent frames, which brought about the multi-target tracking. Simulation results show that the algorithm can not only track different scenarios targets, but also count the number of targets and display target trajectory rapidly and stably. Compared with the particle filter and other metheds, tracking is more smooth, thus the reliability of the tracking is improved.
Keywords:multi-target detection  multi-target tracking  Kalman filter  BP neural netwok
本文献已被 万方数据 等数据库收录!
点击此处可从《红外与激光工程》浏览原始摘要信息
点击此处可从《红外与激光工程》下载免费的PDF全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号