没有合适的资源?快使用搜索试试~ 我知道了~
首页深度视频驱动的视点与时间不变动作识别新方法
深度视频驱动的视点与时间不变动作识别新方法
0 下载量 195 浏览量
更新于2024-08-26
收藏 93KB PDF 举报
本文主要探讨了"基于深度视频的视点和时不变动作识别"这一研究领域。近年来,在传统的RGB视频的人体动作识别(HAR)方面,手工设计特征的方法进展相对有限。然而,随着低成本深度相机的出现,它们在动作识别任务中提供了额外的信息优势。与RGB视频相比,深度视频序列对光线变化更不敏感,并且在诸如分割和活动识别等视觉任务中更具区分度。 作者们针对这个情况,提出了一种有效且直观的深度视频动作识别方法,特别关注骨骼关节信息。首先,他们通过分析深度视频中的关节,计算出三个捕捉关节之间角度和位置信息的特征向量。这些特征向量包含了关于人体姿态的关键信息,能够有效地反映动作的动态特性。 接着,这三组特征向量被分别输入到三个独立的支持向量机(SVM)分类器中。SVM作为非线性分类模型,其在处理高维数据和小型样本集上表现出色,能够有效地从复杂的骨骼运动模式中区分不同的动作类别。这种方法利用深度视频的独特优势,减少了对光照条件的依赖,提高了动作识别的鲁棒性。 最后,通过将多个特征向量和SVM分类器结合起来,该方法能够实现对视点不变性和时间不变性的动作识别,这意味着即使在不同视角或时间点,也能准确地识别出相同的动作类型。这种技术对于许多实际应用,如智能家居、监控系统以及虚拟现实交互等,具有重要的价值和潜力。 总结来说,这篇文章的核心贡献在于开发了一种新颖的深度视频动作识别框架,通过结合深度信息的骨骼特征和SVM分类器,实现了对光照变化和视角变化的稳健识别,为提高RGB视频动作识别领域的性能提供了一种新的可能。
资源详情
资源推荐
An effective view and time-invariant action
recognition method based on depth videos
Zhi Liu
1
, Xin Feng
1
, Yingli Tian
2
1
College of Computer Science and Engineering, Chongqing University of Technology, Chongqing, 400050, China
2
Department of Electrical Engineering, The City College of New York, New York, NY 10031, USA
liuzhi@cqut.edu.cn, xfeng@cqut.edu.cn, ytian@ccny.cuny.edu
Abstract—Little progress has been achieved in hand-crafted
feature based human action recognition (HAR) for RGB videos
in recent years. The emergence of low price depth camera
presents more information for action recognition. Compared
to RGB videos, depth video sequences are more insensitive to
light changes and more discriminative in many vision tasks
such as segmentation and activity recognition. In this paper,
we propose an effective and straightforward HAR method by
using skeleton joints information of the depth sequence. First,
we calculate three feature vectors which capture angle and
position information between joints. Then, the obtained vectors
are used as the inputs of three separate support vector machine
(SVM) classifiers. Finally, the action recognition is conducted
by fusing the SVM classification results. Our features are view-
invariant because the extracted vectors contain only angle and
normalized position information based on joint coordinates. By
normalizing action videos with different temporal lengths to a
fixed size using interpolation, the extracted features have the same
dimension for different videos and can still keep the principal
movement patterns which make the proposed method time-
invariant. Experimental results demonstrate that our method
performs comparable results on the UTKinect-Action3D dataset,
and is more efficient and simpler than state-of-the-art methods.
I. INTRODUCTION
HAR plays an important role in many applications such
as video surveillance, human-computer interaction, video re-
trieval, etc. In past several years, the progress on various
visual recognition tasks has been based mostly on hand-
crafted features including scale-invariant feature transform
(SIFT) [1], histograms of oriented gradient (HOG) [2], motion
history image (MHI) [3] etc. However, most of the canonical
visual recognition algorithms just build ensemble systems and
employee minor variants of successful methods, it is generally
acknowledged that progress has been slow in recent years [4].
Fortunately, the low-cost depth camera promotes researchers
reconsider problems of image processing and computer vision
[5]. Different from RGB camera which captures color and
texture information, depth camera records depth information
with the geometric and skeleton joints information. In addition,
depth camera is insensitive to light changes and more discrim-
inative than color and texture features in many problems such
as segmentation and activity recognition. In this paper, we
propose an effective and straightforward HAR method by only
utilizing skeleton joints information. The proposed method
extracts angle and normalized position information to form
feature vectors from skeleton joint coordinates, which make
it view-invariant. By normalizing action videos with differ-
ent lengths to a fixed size using interpolation, the extracted
features have the same dimension for different video and
keep principal movement patterns which make the proposed
method time-invariant. Experimental results demonstrate that
our method performs comparable results on the UTKinect-
Action3D dataset but is more efficient and simpler than the
state-of-the-art methods The key contributions of this work
are summarized as follows:
1) We propose an effective and simple method for action
recognition just using skeleton joints information for
depth video sequences. Experimental results demon-
strate that our proposed method is time and view-
invariant.
2) Two different hand-crafted joint feature vectors which
are Hip center based vector (HCBV) and angle vector
(AV) are proposed. Pairwise relative position [6] vector
(PRPV) is improved.
3) By fusing classification results from three hand-crafted
features, the recognition accuracy of the proposed
method is comparable to and more efficient and simpler
than the state-of-the-art methods.
The remainder of this paper is organized as follows: Section
II reviews the related work. Section III presents the details
of three hand-crafted features. The experimental results and
discussions are presented in Section IV. Finally, Section V
concludes the paper.
978-1-4673-7314-2/15/$31.00 ©2015 IEEE IEEE VCIP 2015
下载后可阅读完整内容,剩余3页未读,立即下载
weixin_38671048
- 粉丝: 4
- 资源: 870
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
- 社交媒体营销激励优化策略研究
- 终端信息查看工具:qt框架下的输出强制抓取
- MinGW Win32 C/C++ 开发环境压缩包快速入门指南
- STC8G1K08 PWM模块实现10K频率及易改占空比波形输出
- MSP432电机驱动编码器测路程方法解析
- 实现动静分离案例的css/js/img文件指南
- 爱心代码五种:高效编程的精选技巧
- MATLAB实现广义互相关时延估计GCC的多种加权方法
- Hive CDH Jar包下载:免费获取Hive JDBC驱动
- STC8G单片机实现EEPROM及MODBUS-RTU协议
- Java集合框架面试题精讲
- Unity游戏设计与开发资源全集
- 探索音乐盒.zip背后的神秘世界
- Matlab自相干算法GUI界面设计及仿真
- STM32智能小车PID算法实现资料
- Python爬虫实战:高效爬取百度贴吧信息
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功