基于核化相关滤波器的目标跟踪技术研究及应用系统
发布时间：2019-02-14 16:38
【摘要】：基于视频的目标跟踪技术是指使用计算机对视频序列中感兴趣目标进行持续跟踪从而得到目标运动轨迹等信息。目标跟踪技术在安防视频监控、视频文件压缩、人机交互、智能交通等领域都有广泛的应用。然而,由于视频背景环境的复杂性和目标在运动过程中经常会改变外观和形状等因素的影响,导致对目标实现准确的跟踪是一件非常困难的事情。本文主要对运动目标检测与跟踪算法在目标尺度变化、目标受到其它物体遮挡等场景下存在的一些问题进行研究。本文的主要研究工作内容如下:1、单目标跟踪方面。本文首先对核化相关滤波器(Kernelized Correlation Filter,KCF)跟踪算法进行了深入的研究并提出一种遮挡检测算法判定所跟踪目标是否受到其它物体的遮挡干扰,进而使用卡尔曼滤波器对受到遮挡干扰的目标进行位置预测。通过实验可知,加入遮挡检测后的核化相关滤波器跟踪算法较好地解决了目标在线性运动下受到其它物体遮挡干扰的问题;然后,本文使用相关滤波器(CorrelationFilter,CF)和图像尺度金字塔计算目标的尺度大小,较好地解决了目标在运动过程中存在的尺度变化问题。2、多目标跟踪方面。本文首先在运动目标检测算法中融入深度图像,较好解决了运动目标的阴影干扰。接着,本文提出了一种基于核化相关滤波器(KCF)跟踪算法和运动目标检测的多目标跟踪算法。在目标之间没有相互遮挡情况下,使用运动目标检测算法得到运动目标区域,在运动目标区域上进行人脸检测后使用卡尔曼滤波器和数据关联技术生成多目标轨迹;一旦判定多个目标将要交互,对每个目标都使用KCF跟踪算法进行独立跟踪。实验结果表明,提出的多目标跟踪算法能够较好地处理多目标之间存在的遮挡问题。3、最后本文设计一个基于核化相关滤波器目标跟踪算法的应用系统,并实现了基于视频目标跟踪技术的视频监控网络微信报警子系统、跑步视频分析子系统和多目标人数统计子系统。验证了本文所提出的改进目标跟踪算法在现实生活场景中的可行性。
[Abstract]:Video based target tracking technology refers to the continuous tracking of objects of interest in video sequences using computers to obtain information such as the moving trajectory of the target. Target tracking technology has been widely used in security video surveillance, video file compression, human-computer interaction, intelligent transportation and other fields. However, due to the complexity of the video background environment and the influence of factors such as changing the appearance and shape of the target in the process of moving, it is very difficult to track the target accurately. In this paper, some problems in moving target detection and tracking algorithms are studied, such as the change of the target scale and the occlusion of the target by other objects. The main work of this paper is as follows: 1. Single target tracking. In this paper, we study the (Kernelized Correlation Filter,KCF (Kernelized correlation filter) tracking algorithm and propose an occlusion detection algorithm to determine whether the target is affected by other objects. Furthermore, Kalman filter is used to predict the position of the target affected by occlusion interference. The experimental results show that the kernel correlation filter tracking algorithm with occlusion detection can solve the problem that the target is blocked by other objects under linear motion. Then, the correlation filter (CorrelationFilter,CF) and the image scale pyramid are used to calculate the size of the target, which can solve the problem of the scale change in the moving process. 2. Multi-target tracking. Firstly, the depth image is incorporated into the moving target detection algorithm, which solves the shadow interference of moving target. Then, a multi-target tracking algorithm based on Kernel correlation filter (KCF) and moving target detection is proposed. When there is no mutual occlusion between targets, moving target detection algorithm is used to obtain moving target region, and Kalman filter and data association technique are used to generate multi-target trajectory after face detection in moving target area. Once it is determined that multiple targets will interact, each target is tracked independently using the KCF tracking algorithm. Experimental results show that the proposed multi-target tracking algorithm can deal with the occlusion problem between multiple targets. 3. Finally, this paper designs an application system based on the Kernel correlation filter target tracking algorithm. The video surveillance network WeChat alarm subsystem, running video analysis subsystem and multi-target statistics subsystem are implemented based on video target tracking technology. The feasibility of the proposed improved target tracking algorithm in real life scenarios is verified.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
[Abstract]:Video based target tracking technology refers to the continuous tracking of objects of interest in video sequences using computers to obtain information such as the moving trajectory of the target. Target tracking technology has been widely used in security video surveillance, video file compression, human-computer interaction, intelligent transportation and other fields. However, due to the complexity of the video background environment and the influence of factors such as changing the appearance and shape of the target in the process of moving, it is very difficult to track the target accurately. In this paper, some problems in moving target detection and tracking algorithms are studied, such as the change of the target scale and the occlusion of the target by other objects. The main work of this paper is as follows: 1. Single target tracking. In this paper, we study the (Kernelized Correlation Filter,KCF (Kernelized correlation filter) tracking algorithm and propose an occlusion detection algorithm to determine whether the target is affected by other objects. Furthermore, Kalman filter is used to predict the position of the target affected by occlusion interference. The experimental results show that the kernel correlation filter tracking algorithm with occlusion detection can solve the problem that the target is blocked by other objects under linear motion. Then, the correlation filter (CorrelationFilter,CF) and the image scale pyramid are used to calculate the size of the target, which can solve the problem of the scale change in the moving process. 2. Multi-target tracking. Firstly, the depth image is incorporated into the moving target detection algorithm, which solves the shadow interference of moving target. Then, a multi-target tracking algorithm based on Kernel correlation filter (KCF) and moving target detection is proposed. When there is no mutual occlusion between targets, moving target detection algorithm is used to obtain moving target region, and Kalman filter and data association technique are used to generate multi-target trajectory after face detection in moving target area. Once it is determined that multiple targets will interact, each target is tracked independently using the KCF tracking algorithm. Experimental results show that the proposed multi-target tracking algorithm can deal with the occlusion problem between multiple targets. 3. Finally, this paper designs an application system based on the Kernel correlation filter target tracking algorithm. The video surveillance network WeChat alarm subsystem, running video analysis subsystem and multi-target statistics subsystem are implemented based on video target tracking technology. The feasibility of the proposed improved target tracking algorithm in real life scenarios is verified.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
【参考文献】
相关期刊论文 前2条
1 蒋恋华;甘朝晖;蒋e，
本文编号：2422395
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2422395.html
