没有合适的资源?快使用搜索试试~ 我知道了~
首页Real Time Head Pose Estimation with Random Regression Forests
Real Time Head Pose Estimation with Random Regression Forests
4星 · 超过85%的资源 需积分: 6 18 下载量 180 浏览量
更新于2023-03-03
评论
收藏 1.13MB PDF 举报
Real Time Head Pose Estimation with Random Regression Forests是cvpr2011的优秀论文,这里提供大家下载和学习,另外有源代码是matlab的
资源详情
资源评论
资源推荐
Real Time Head Pose Estimation with Random Regression Forests
Gabriele Fanelli
1
Juergen Gall
1
Luc Van Gool
1,2
1
BIWI, ETH Zurich
2
ESAT-PSI / IBBT, KU Leuven
{fanelli,gall,vangool}@vision.ee.ethz.ch vangool@esat.kuleuven.be
Abstract
Fast and reliable algorithms for estimating the head pose
are e ssential for many applications and higher-level face
analysis tasks. We address the problem of head pose esti-
mation from depth data, which can be captured u sing the
ever mo re affordable 3D sensing technologies available to -
day. To achieve robustness, we formulate pose estimation
as a regression problem. While detecting specific face parts
like the nose is sensitive to occlusions, learning the regres-
sion on rather generic surfac e patches requires enormous
amount of training data in order to a chieve accurate esti-
mates. We propose to use random regression forests for the
task at hand, given their capability to handle large training
datasets. Moreover, we synthesize a great amo unt of anno-
tated training data using a statistical model of the human
face. In our experiments, we show that our approach can
handle real data presenting large pose changes, partial oc-
clusions, and facial expressions, even though it is trained
only on synthetic neutral face data. We ha v e thoroughly
evaluated our system on a publicly available database on
which we achieve state-of-the-art performance without hav-
ing to resort to the graphics card.
1. Introduction
Automatic and robust algorithms for head pose estima-
tion can be beneficial to many re al life applications. Accu-
rately localizing the head and its orientation is either the ex-
plicit goal of systems like human-computer interfaces (e.g.,
reacting to the user’s head movements), or a necessary pre-
processing step for further analysis, such as identificatio n
or facial expression recognition . Due to its relevance and to
the challenges posed by the problem, there has been consid-
erable effort in the computer vision community to develop
fast and reliable algorithms for head pose estimation.
Methods relying solely on standard 2D images face ser i-
ous pro blems, notably illumination changes and textureless
face r egio ns. Given the recent development and availabil-
ity of 3D sensing technologies, which are becoming ever
more affordable and reliable, the additional depth informa-
Figure 1. Real time head pose estimation example.
tion can finally allow us to overcome some of the prob-
lems inherent of methods based on 2D data. However, ex-
isting depth-based methods either need manual initializa-
tion, cannot handle large pose variations, or are not real-
time. An exception are approaches like the one presented
by [4], where the au thors achieve real-time performance by
exploiting th e massive parallel processing power of a GPU.
Their approach relies on a geome tric descriptor which pro-
vides nose location hypotheses which are then compare d
to a large num ber of renderin gs of a generic face template,
done in p a rallel on the GPU. The fast computatio n time
reported is only achievable provid e d tha t specific gra phics
hardware is available.
GPUs, however, present a very high power consump-
tion which limits their use for certain kinds of applica tion.
Hence, we propose an approach for 3D head pose estima-
tion which doe s not rely on specific graphics hardware and
which can be tuned to achieve the desired trade-off between
accuracy an d computation cost, which is particularly useful
when resources are limited by the application. We formu-
late the problem as a regression, estimating th e head pose
parameters direc tly from the depth data. The regression
is implemented within a random forest framework [2, 10],
learning a mapping from simple d epth features to a prob-
abilistic estimation of real-valued parameters such as 3D
nose coordinates and head rotation angles. Since random
forests (as any regressor ) need to be trained o n labeled data
and the accuracy depends on the amount of training, data
617
abyssecho
- 粉丝: 1
- 资源: 15
上传资源 快速赚钱
- 我的内容管理 收起
- 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
会员权益专享
最新资源
- 2022年中国足球球迷营销价值报告.pdf
- 房地产培训 -营销总每天在干嘛.pptx
- 黄色简约实用介绍_汇报PPT模板.pptx
- 嵌入式系统原理及应用:第三章 ARM编程简介_3.pdf
- 多媒体应用系统.pptx
- 黄灰配色简约设计精美大气商务汇报PPT模板.pptx
- 用matlab绘制差分方程Z变换-反变换-zplane-residuez-tf2zp-zp2tf-tf2sos-sos2tf-幅相频谱等等.docx
- 网络营销策略-网络营销团队的建立.docx
- 电子商务示范企业申请报告.doc
- 淡雅灰低面风背景完整框架创业商业计划书PPT模板.pptx
- 计算模型与算法技术:10-Iterative Improvement.ppt
- 计算模型与算法技术:9-Greedy Technique.ppt
- 计算模型与算法技术:6-Transform-and-Conquer.ppt
- 云服务安全风险分析研究.pdf
- 软件工程笔记(完整版).doc
- 电子商务网项目实例规划书.doc
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
评论2