视频序列中运动目标检测与跟踪算法研究

发布时间：2018-03-27 07:07

本文选题：运动目标检测与跟踪　切入点：高斯混合模型　出处：《长春工业大学》2017年硕士论文

【摘要】：视频序列中运动目标的检测与跟踪一直是计算机视觉领域研究中倍受关注的热门课题,并被广泛应用于精确武器制导、智能监控、人机交互、智能机器人等军事和日常生活中。这一课题不仅涉及了多个学科领域,而且其应用的实际场景也是不尽相同。虽然人们针对这一课题进行了大量的研究工作,但是还是存在着一些重要问题尚未解决。本论文不仅对前人的检测和跟踪算法进行了深入的分析和研究,而且在某些方面给出了自己的改进算法,使其更趋于完善。在运动目标检测方面,本论文分析了三种常用的运动目标检测方法,即:光流法、帧间差分法、背景减除法。着重对基于高斯混合模型的背景减除法做了分析和研究,此方法主要包括前景检测、像素级后处理、区域分析、区域级后处理、特征提取五个环节。通过分析和研究发现传统的基于高斯混合模型的背景减除法,在前景检测环节中存在着缺陷,算法收敛速度慢,并且算法需要大量的存储空间。于是在前景检测算法的基础上,对于该算法的具体实现步骤进行了改进,给出了一种新的高斯混合模型初始化方法,利用在线K-均值聚类的方法对高斯混合模型进行初始化,同时对模型更新方法做了进一步改进和优化,对匹配准则和新高斯分布生成准则做了改进。通过前景检测的仿真实验发现改进算法不但提高了检测算法的收敛速度,而且具有很好的稳定性,此外,从算法运行时所占用的存储空间上比较,实验证明节约了近一半的存储空间。在运动目标跟踪方面,本论文对基于Kalman滤波的运动目标跟踪算法和采用颜色直方图作为跟踪特征的Camshift跟踪算法做了深入的研究和分析。通过分析发现,当运动目标周围存在着与其具有相似颜色特征的大面积干扰物时,Camshift跟踪算法不能准确的跟踪运动目标。针对这一问题,本文给出了一种将Camshift跟踪算法和基于Kalman滤波的跟踪算法两者相结合的新的运动目标跟踪算法,通过实验证明,给出的新算法能够有效的解决大面积颜色干扰的问题。
[Abstract]:Detection and tracking of moving targets in video sequences has been a hot topic in the field of computer vision, and has been widely used in precision weapon guidance, intelligent monitoring, human-computer interaction. In military and daily life, such as intelligent robots, this subject not only involves many disciplines, but also has different application scenarios. Although people have done a lot of research on this subject, However, there are still some important problems that remain unsolved. This paper not only analyzes and studies the previous detection and tracking algorithms, but also gives its own improved algorithm in some aspects. In the aspect of moving target detection, this paper analyzes three common moving target detection methods, namely: optical flow method, inter-frame difference method, Background subtraction method. The background subtraction method based on Gao Si's mixed model is analyzed and studied. This method mainly includes foreground detection, pixel level post processing, region analysis, region level post processing, Through analysis and research, it is found that the traditional background subtraction method based on Gao Si's mixed model has some defects in foreground detection, and the convergence speed of the algorithm is slow. And the algorithm needs a lot of storage space. Therefore, based on the foreground detection algorithm, the implementation steps of the algorithm are improved, and a new initialization method of Gao Si hybrid model is proposed. The online K-means clustering method is used to initialize Gao Si's mixed model, and the model updating method is further improved and optimized. The matching criterion and the new Gao Si distribution generation criterion are improved. The simulation results of foreground detection show that the improved algorithm not only improves the convergence speed of the detection algorithm, but also has good stability. Compared with the storage space occupied by the algorithm, the experimental results show that nearly half of the storage space is saved. In this paper, the moving target tracking algorithm based on Kalman filter and the Camshift tracking algorithm based on color histogram are studied and analyzed. The Camshift tracking algorithm can not track moving targets accurately when there is a large area of disturbance with similar color characteristics around the moving target. In this paper, a new moving target tracking algorithm which combines Camshift tracking algorithm and Kalman filter tracking algorithm is presented. The experimental results show that the new algorithm can effectively solve the problem of large area color interference.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】