CASIA OpenIR  > 毕业生  > 硕士学位论文
基于视觉的运动检测与跟踪算法研究
其他题名Study on Vision-based Motion Detection and Tracking
王萍
学位类型工学硕士
导师胡包钢
2002-05-01
学位授予单位中国科学院研究生院
学位授予地点中国科学院自动化研究所
学位专业模式识别与智能系统
关键词视觉跟踪 物体定位 跟踪初始化 轮廓跟踪 区域跟踪 颜色特征量 图像相减 Visual Tracking Object Localization Tracking Initialization Contour-based Tracing Region-based Tracking Color Feature Image Subt
摘要视觉跟踪不论是算法研究上,还是实际跟踪系统的建立上,近十几年来都取得了长足的发展。它是计算机视觉领域中的一个重要问题,有重大的军事和民用实际意义。随着计算机硬件速度的提高,设计实时的稳定的跟踪系统已成为可能,每年都有许多这方面的研究论文发表。本文讨论了视觉跟踪中的如下几个问题:跟踪的初始化:跟踪算法的比较研究:静止背景下彩色图像序列的运动检测与跟踪。目前的跟踪算法中,很多都不能实现自动检测目标并跟踪,一般的目标位置初始化都是靠手动完成。本文着重讨论了基于概率统计模型的在单幅图像中对目标进行定位的方法,并通过实验说明了该方法的有效性。与跟踪算法研究快速发展不相适应的是对跟踪算法进行评估的系统性研究。跟踪算法比较研究的难点在于目前的跟踪都是面向特定应用的,没有通用的跟踪理论,没有公认的标准的测试序列,因此很难制定一个通用的评价标准适用于所有的算法,但是我们需要有这样的评价指标作为选择合适算法的参考。本文尝试做一些跟踪算法比较方面的工作,从理论及实验上比较了基于轮廓的跟踪算法和基于区域信息的跟踪算法。其中基于轮廓的跟踪算法以近年提出的CONDENSATION算法为代表,基于区域的跟踪算法以相关模板匹配法为代表。理论上比较了算法的假设条件、所需先验知识、计算量大小等方面;实验上对两个算法的速度及鲁棒性进行测试,衡量指标有每帧平均处理时间及正确跟踪帧数。这些评价指标为人们选择特定环境下的跟踪算法提供了综合的客观的参考依据。结论指出,相关模板匹配法适用于背景简单、运动目标在跟踪过程中变化不大的情况;CONDENSATION算法在背景杂乱、运动过程较复杂的情况下工作表现优秀。颜色信息是被广泛应用的图像底层可视化特征,它在拉伸、旋转、透视变换和有遮挡的情况下都能保持鲁棒的特性。本文提出了对彩色图像序列在静止背景的运动检测与跟踪的简单有效的方法。通过比较各颜色特征量的大小来判断图像中的运动区域,并对颜色特征量的作用进行定量分析。其中对于参考图像的选择也是本方法中的一个创新点,它避开了背景图像的学习问题,使得人们利用图像相减技术时更具灵活性。
其他摘要Visual tracking has made significant progresses both in algorithm design and practical tracking system development in recent years. It is one of the most difficult problems in computer vision and the research in visual tracking will bring great meaning to national defense and everyday applications. With the high processing speed in computer hardware, real-time tracking has the possibility to come into reality. There are many papers on such topics published every year. This dissertation analyzes the following problems in visual tracking area: initialization of tracking; algorithms comparison and detecting and tracking moving objects from a static background scene using color images. Current algorithms seldom pay attention to the initialization problem in tracking. Most of them do this by hand, while tracking automatically is on the demand. This dissertation discusses an object localization method based on statistical model and shows its effect through experiments. What not accommodates to the active research in tracking algorithms is the lack of systematical evaluation of tracking results. The evaluation difficulties lies in the fact that all the tracking algorithms are working under given tasks and there are no unified theory of visual tracking and no standard testing image sequences. So it is hard to establish a general evaluation guideline for all the algorithms, but people need such reference to choose proper method for a certain environment. Some comparison work of tracking algorithms has been done between region-based and contour-based methods in this dissertation. It focuses on the template-based region tracking method and CONDENSATION algorithm. Analysis is made in algorithms' assumption, prior information, computation and running speed. Conclusion is that template- based region tracking works well in simple background with the target being roughly constant from frame to frame, while CONDENSATION algorithm is especially useful when dealing with cluttered background and unknown change in motion. Color information is the most intensively used visual feature in virtue of its strong correlation with the underlying image objects or scenes. Compared to other low-level visual information, color is more robust with respect to scaling, orientation, perspective and occlusion of images. A new algorithm of detecting and tracking moving objects from a static background scene using color images is proposed. It is based on the comparison of color features to decide moving areas in the image. The effects of color features are analyzed in this dissertation. Another innovation in this algorithm is the choice of reference background image, since it doesn't require background learning process.
馆藏号XWLW641
其他标识符641
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/6799
专题毕业生_硕士学位论文
推荐引用方式
GB/T 7714
王萍. 基于视觉的运动检测与跟踪算法研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2002.
条目包含的文件
条目无相关文件。
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[王萍]的文章
百度学术
百度学术中相似的文章
[王萍]的文章
必应学术
必应学术中相似的文章
[王萍]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。