大工至善|大学至真分享 http://blog.sciencenet.cn/u/lcj2212916

博文

[转载]【信息技术】【2014.05】视频图像中的目标检测与跟踪

已有 173 次阅读 2020-6-15 17:37 |系统分类:科研笔记|文章来源:转载

本文为印度Rourkela国立技术研究院(作者:Rajkamal kishor Gupta)的硕士论文,共46页。

 

近年来,由于采用成本较低且技术优越的摄像设备,其图像质量得到了迅速的提高,因此拍摄高质量、大尺寸的图像变得非常容易。视频是具有固定时间间隔的连续图像的集合。因此,当场景随时间变化时,视频可以提供有关对象的更多信息。手动处理视频是不可能的。所以我们需要一个自动化的设备来处理这些视频。

 

本文对视频中的目标跟踪进行了一次尝试。已经开发了许多算法和技术来自动监视视频文件中的对象。目标检测与跟踪是计算机视觉领域的一项具有挑战性的任务。视频分析的基本步骤主要有三个:从运动目标中检测出感兴趣的目标,在连续帧中跟踪感兴趣的目标,分析目标的运动轨迹以了解其行为。简单的目标检测将像素级的静态背景帧与当前视频帧进行比较,该领域的现有方法尝试检测视频帧中的感兴趣对象。目标跟踪的主要困难之一是选择合适的特征和模型来识别、跟踪视频中感兴趣的目标。一些常见的选择是根据类别选择合适的特征,视觉对象包括强度、形状、颜色和特征点。本文研究了基于彩色pdfmean-shift跟踪、基于强度和运动的光流跟踪、基于尺度不变局部特征点的SIFT跟踪。实验初步结果表明,所采用方法能够对平移、旋转、部分遮挡和形变目标进行跟踪。

 

In recent days, capturing images with highquality and good size is so easy because of rapid improvement in quality ofcapturing device with less costly but superior technology. Videos are acollection of sequential images with a constant time interval. So video canprovide more information about our object when scenarios are changing withrespect to time. Therefore, manually handling videos are quite impossible. Sowe need an automated devise to process these videos. In this thesis one suchattempt has been made to track objects in videos. Many algorithms andtechnology have been developed to automate monitoring the object in a videofile. Object detection and tracking is a one of the challenging task incomputer vision. Mainly there are three basic steps in video analysis:Detection of objects of interest from moving objects, Tracking of thatinterested objects in consecutive frames, and Analysis of object tracks tounderstand their behavior. Simple object detection compares a static backgroundframe at the pixel level with the current frame of video. The existing methodin this domain first tries to detect the interest object in video frames. Oneof the main difficulties in object tracking among many others is to choosesuitable features and models for recognizing and tracking the interested objectfrom a video. Some common choice to choose suitable feature to categories, visualobjects are intensity, shape, color and feature points. In this thesis, westudied about mean shift tracking based on the color pdf, optical flow trackingbased on the intensity and motion; SIFT tracking based on scale invariant localfeature points. Preliminary results from experiments have shown that theadopted method is able to track targets with translation, rotation, partialocclusion and deformation.

 

1. 引言

2. 目标检测与跟踪

3. 特征提取方法

4. 实验结果

5. 结论与未来工作展望


更多精彩文章请关注公众号:205328s611i1aqxbbgxv19.jpg




http://blog.sciencenet.cn/blog-69686-1237992.html

上一篇:[转载]【计算机科学】【2018】组织学图像分类的深度学习结构
下一篇:[转载]【计算机科学】【2017.05】卷积神经网络结构的研究

0

该博文允许注册用户评论 请点击登录 评论 (0 个评论)

数据加载中...

Archiver|手机版|科学网 ( 京ICP备07017567号-12 )

GMT+8, 2020-9-21 17:22

Powered by ScienceNet.cn

Copyright © 2007- 中国科学报社

返回顶部