视频目标跟踪改进算法的研究

发布时间：2018-06-06 09:30

本文选题：视频目标跟踪 + Camshift　；参考：《江南大学》2017年硕士论文

【摘要】：作为计算机领域的一个重要研究课题,视频目标跟踪技术在军事、监控、汽车导航以及医疗领域等方面有非常广阔的发展前景。随着计算机技术的不断发展,计算机的功能不断得到强化,为了使计算机能模拟人类视觉神经系统的运动功能,视频目标跟踪技术应运而生。近年来,大量的研究学者对视频目标跟踪技术进行了深入研究,并提出了许多新颖的理论和相关跟踪算法。但是,由于存在杂乱背景、局部遮挡、光照变化以及运动模糊等因素的干扰,视频目标跟踪技术依然存在很多问题。因此开发出具有高鲁棒性、高准确性以及实用性的视频目标跟踪算法仍然面临着挑战,并且具有广阔的研究前景和实用价值。本文在研究经典视频目标跟踪算法的基础上,结合数据挖掘领域的知识,提出三个改进算法用于提高跟踪算法的跟踪性能。(1)为了提高传统Camshift算法在视频跟踪中的鲁棒性,提出一种基于黑洞原理和Camshift算法的跟踪算法,该算法利用黑洞原理对视频序列的特征颜色分量权值进行选择,根据最优颜色分量权重重构图像,将重构图像转化到HSV空间,计算HSV空间中H分量的颜色概率图,利用颜色概率图的一阶矩以及零阶矩得到目标的中心,从而提高跟踪算法的精度。(2)针对目标跟踪过程中目标物体的外观变化影响,提高目标跟踪的鲁棒性,提出一种改进的稀疏外观模型目标跟踪算法,该算法通过从目标模板几何中选取主要目标模板来降低目标模板数量。算法首先通过学习分类器得到目标特征,然后用黑洞原理获取的模板字典表示目标,采用高斯分布运动模型获取目标样本,在贝叶斯框架下根据观测模型获取最优目标位置实现跟踪。(3)为了提高遮挡情况下目标跟踪的精度,提出一种基于主成分分析的目标跟踪算法,该算法利用主成分分析对目标模板进行更新,并对目标模板进行优化处理,减少模板数量,提高跟踪的精度和减少计算量。
[Abstract]:As an important research topic in the field of computer, video target tracking technology has a very broad prospect in military, surveillance, vehicle navigation and medical field. With the development of computer technology, the function of computer has been strengthened constantly. In order to make the computer simulate the motion function of human visual nervous system, video target tracking technology emerges as the times require. In recent years, a large number of researchers have carried out in-depth research on video target tracking technology, and put forward many novel theories and related tracking algorithms. However, due to the interference of clutter background, local occlusion, illumination change and motion blur, there are still many problems in video target tracking technology. Therefore, the development of video target tracking algorithm with high robustness, high accuracy and practicability still faces challenges, and has broad research prospects and practical value. Based on the research of classical video target tracking algorithm and the knowledge of data mining, this paper proposes three improved algorithms to improve the tracking performance of the tracking algorithm. In order to improve the robustness of the traditional Camshift algorithm in video tracking, this paper proposes three improved algorithms to improve the tracking performance of the tracking algorithm. A tracking algorithm based on black hole principle and Camshift algorithm is proposed. The black hole principle is used to select the weights of feature color components of video sequences. According to the optimal weight of color components, the reconstructed image is reconstructed and transformed into HSV space. The color probability map of H component in HSV space is calculated. The center of the target is obtained by using the first-order moment and the zero-order moment of the color probability map, so as to improve the accuracy of the tracking algorithm. To improve the robustness of target tracking, an improved sparse appearance model target tracking algorithm is proposed. The algorithm reduces the number of target templates by selecting the main target template from the geometry of the target template. Firstly, the target feature is obtained by learning classifier, then the target is represented by the template dictionary obtained by the black hole principle, and the target sample is obtained by using the Gao Si distributed motion model. In order to improve the accuracy of target tracking under occlusion, a tracking algorithm based on principal component analysis (PCA) is proposed. The algorithm uses principal component analysis to update the target template, and optimizes the target template to reduce the number of templates, improve the tracking accuracy and reduce the amount of calculation.
【学位授予单位】：江南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】