Reinforcement Learning & AlphaGo

Similar documents
(JBE Vol. 23, No. 2, March 2018) (Special Paper) 23 2, (JBE Vol. 23, No. 2, March 2018) ISSN

High Resolution Disparity Map Generation Using TOF Depth Camera In this paper, we propose a high-resolution disparity map generation method using a lo

(......).hwp

Software Requirrment Analysis를 위한 정보 검색 기술의 응용

<4D F736F F D20B1E2C8B9BDC3B8AEC1EE2DC0E5C7F5>

<4D F736F F D20C3D6BDC C0CCBDB4202D20BAB9BBE7BABB>

다중 곡면 검출 및 추적을 이용한 증강현실 책

07.045~051(D04_신상욱).fm

0914ä¿òÇÕº».PDF

Ch 1 머신러닝 개요.pptx

2 : (Seungsoo Lee et al.: Generating a Reflectance Image from a Low-Light Image Using Convolutional Neural Network) (Regular Paper) 24 4, (JBE

<313120C0AFC0FCC0DA5FBECBB0EDB8AEC1F2C0BB5FC0CCBFEBC7D15FB1E8C0BAC5C25FBCF6C1A42E687770>

<4D F736F F D20C3D6BDC C0CCBDB4202D20BAB9BBE7BABB>

박선영무선충전-내지

SchoolNet튜토리얼.PDF

THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE Jul.; 29(7),

DIY 챗봇 - LangCon

Artificial Intelligence: Assignment 6 Seung-Hoon Na December 15, Sarsa와 Q-learning Windy Gridworld Windy Gridworld의 원문은 다음 Sutton 교재의 연습문제

(JBE Vol. 22, No. 2, March 2017) (Regular Paper) 22 2, (JBE Vol. 22, No. 2, March 2017) ISSN

보고싶었던 Deep Learning과 OpenCV를이용한이미지처리과정에대해공부를해볼수있으며더나아가 Deep Learning기술을이용하여논문을작성하는데많은도움을받을수있으며아직배우는단계에있는저에게는기존의연구를따라해보는것만으로도큰발전이있다고생각했습니다. 그래서이번 DSP스마

PowerPoint 프레젠테이션

12권2호내지합침

PowerPoint 프레젠테이션

Gray level 변환 및 Arithmetic 연산을 사용한 영상 개선

Multi-pass Sieve를 이용한 한국어 상호참조해결 반-자동 태깅 도구

PART

Part Part

£01¦4Àå-2

½ºÅ丮ÅÚ¸µ3_³»Áö

272*406OSAKAÃÖÁ¾-¼öÁ¤b64ٽÚ

I

Electronics and Telecommunications Trends 인공지능을이용한 3D 콘텐츠기술동향및향후전망 Recent Trends and Prospects of 3D Content Using Artificial Intelligence Technology

第 1 節 組 織 11 第 1 章 檢 察 의 組 織 人 事 制 度 등 第 1 項 大 檢 察 廳 第 1 節 組 대검찰청은 대법원에 대응하여 수도인 서울에 위치 한다(검찰청법 제2조,제3조,대검찰청의 위치와 각급 검찰청의명칭및위치에관한규정 제2조). 대검찰청에 검찰총장,대

Microsoft PowerPoint - 실습소개와 AI_ML_DL_배포용.pptx

(JBE Vol. 24, No. 2, March 2019) (Special Paper) 24 2, (JBE Vol. 24, No. 2, March 2019) ISSN

PowerPoint 프레젠테이션

방송공학회논문지 제18권 제2호


R을 이용한 텍스트 감정분석

Delving Deeper into Convolutional Networks for Learning Video Representations - Nicolas Ballas, Li Yao, Chris Pal, Aaron Courville arXiv:

BSC Discussion 1

15_3oracle

<4D F736F F D DB1E2BCFAB5BFC7E2BAD0BCAE2DBEF3B1BCC0CEBDC42DC3A4BFF8BCAE2E646F6378>

1. 서 론

3 Gas Champion : MBB : IBM BCS PO : 2 BBc : : /45

19_9_767.hwp

(JBE Vol. 23, No. 5, September 2018) (Regular Paper) 23 5, (JBE Vol. 23, No. 5, September 2018) ISSN

첨 부 1. 설문분석 결과 2. 교육과정 프로파일 169

( 분류및특징 ) 학습방법에따라 1 지도학습 (Supervised 2 비지도 학습 (Unsupervised 3 강화학습 (Reinforcement 으로구분 3) < 머신러닝의학습방법 > 구분 지도학습 (Supervised 비지도학습 (Unsupervised 강화학습 (

. 서론,, [1]., PLL.,., SiGe, CMOS SiGe CMOS [2],[3].,,. CMOS,.. 동적주파수분할기동작조건분석 3, Miller injection-locked, static. injection-locked static [4]., 1/n 그림

DBPIA-NURIMEDIA

Safe and Humanized Bir t h

ȲÀμº Ãâ·Â

< B3E220C3DFB0E8C1BEC7D5C7D0BCFAB4EBC8B820B9D720C1A4B1E2C3D1C8B820BEC8B3BBC0E E31302E E687770>

°í¼®ÁÖ Ãâ·Â

REP - CP - 016, N OVEMBER 사진 요약 25 가지 색상 Surf 를 이용한 사진 요약과 사진 배치 알고리즘 Photo Summarization - Representative Photo Selection based on 25 Color Hi

39호뉴스레터.indd

4 : (Hyo-Jin Cho et al.: Audio High-Band Coding based on Autoencoder with Side Information) (Special Paper) 24 3, (JBE Vol. 24, No. 3, May 2019

PowerPoint 프레젠테이션

09권오설_ok.hwp

<372DBCF6C1A42E687770>

Data Industry White Paper

PowerPoint Presentation

기획 1 서울공대생에게 물었다 글 재료공학부 1, 이윤구 재료공학부 1, 김유리 전기정보공학부 1, 전세환 편집 재료공학부 3, 오수봉 이번 서울공대생에게 물었다! 코너는 특별히 설문조사 형식으로 진행해 보려고 해 요. 설문조사에는 서울대학교 공대 재학생 121명, 비

기관별 공동 Template

MVVM 패턴의 이해

<35335FBCDBC7D1C1A42DB8E2B8AEBDBAC5CDC0C720C0FCB1E2C0FB20C6AFBCBA20BAD0BCAE2E687770>

PCServerMgmt7

푸른21탄소중립행사내지확정

2


À±½Â¿í Ãâ·Â

(JBE Vol. 23, No. 5, September 2018) (Special Paper) 23 5, (JBE Vol. 23, No. 5, September 2018) ISSN

김기남_ATDC2016_160620_[키노트].key

- 2 -

<4D F736F F D20C3D6BDC C0CCBDB4202D20BAB9BBE7BABB>

RM hwp

제1강 인공지능 개념과 역사

Probability Overview Naive Bayes Classifier Director of TEAMLAB Sungchul Choi

Intra_DW_Ch4.PDF

(JBE Vol. 7, No. 4, July 0)., [].,,. [4,5,6] [7,8,9]., (bilateral filter, BF) [4,5]. BF., BF,. (joint bilateral filter, JBF) [7,8]. JBF,., BF., JBF,.

(JBE Vol. 23, No. 2, March 2018) (Special Paper) 23 2, (JBE Vol. 23, No. 2, March 2018) ISSN

광운소식-68호F

Artificial Intelligence: Assignment 3 Seung-Hoon Na November 30, Sarsa와 Q-learning Windy Gridworld Windy gridworld는 (Sutton 교재 연습문제 6.5) 다음

3 : 3D (Seunggi Kim et. al.: 3D Depth Estimation by a Single Camera) (Regular Paper) 24 2, (JBE Vol. 24, No. 2, March 2019)

(JBE Vol. 24, No. 1, January 2019) (Special Paper) 24 1, (JBE Vol. 24, No. 1, January 2019) ISSN 2287-

2015 경제ㆍ재정수첩

untitled

정보기술응용학회 발표

untitled

0125_ 워크샵 발표자료_완성.key

1 : HEVC Rough Mode Decision (Ji Hun Jang et al.: Down Sampling for Fast Rough Mode Decision for a Hardware-based HEVC Intra-frame encoder) (Special P

<65B7AFB4D7B7CEB5E5BCEEBFEEBFB5B0E1B0FABAB8B0EDBCAD5FC3D6C1BE2E687770>


2 차원단위블록정렬을이용한 내용기반이미지매칭 장철진 O 조환규부산대학교컴퓨터공학과 {jin, Content-based image matching based on 2D alignment of unit block tessellation C

서재초등학교 5학년 학교생활

example code are examined in this stage The low pressure pressurizer reactor trip module of the Plant Protection System was programmed as subject for

±è¼ºÃ¶ Ãâ·Â-1

THE JOURNAL OF KOREAN INSTITUTE OF ELECTROMAGNETIC ENGINEERING AND SCIENCE Sep.; 26(10),

정연덕( )-97.PDF

Transcription:

Gait recognition using a Discriminative Feature Learning Approach for Human identification 딥러닝기술및응용딥러닝을활용한개인연구주제발표 이장우 wkddn1108@kist.re.kr 2018.12.07

Overview 연구배경 관련연구 제안하는방법 Reference 2

I. 연구배경 Reinforcement Learning and AlphaGo Part.1 3

연구배경 기존의생체인식방식 범죄현장이나일상생활에서생체인식의어려움이 기존의방식은인식을위한인위적인환경이요구 얼굴인식의경우, 원거리영상에대하여인식률이크게저하 객체의협조가필요홍채인식이나지문인식의경우, 인식대상자가눈이나손가락을가까이접촉시켜야함 생체인증방식에서의걸음걸이인식 사람의보행패턴을이용하여, 특정대상을식별하는방법 인식대상이촬영되기만하면원거리에서객체인식이가능 인식을위해, 객체의협조가필요하지않다. 4

Gait recognition 걸음걸이인식 사람들이걷는방식에따라개개인을구별하는인식방법 거리를두고얻어진영상을통해, 인식을수행 HID(Human Identification at a Distance) 걸음걸이인식을위한특징 1. Model based feature 모션정보, 걸음걸이영상의 optical flow 다리의각도, 키와보폭등의분석을통한특징추출 2. Model free feature 걸음걸이실루엣영상전체를사용하여, 이를하나의패턴으로사용 <Optical flow> <Motion based Optical flow> <silhouette image> 5

인식데이터의특성 Intra-class 동일클래스내부의분산 Inter-class 서로다른클래스간의분산 Intra class 의분산이크고, Inter class 의분산은작을때, 분류가쉬움 6

연구목표 Model free 걸음걸이이미지는환경과각도에따라, 데이터의특징이굉장히불규칙함 데이터의특징을잘추출하는적합한 Feature extract Model 을찾기가어려움 데이터를최소한의방법으로전처리하고, 이를분류기의입력으로사용 End to End Feature 를잘분류하는적합한분류기모델을선정하기어려움 딥러닝을활용하여 Feature 추출과새로운데이터에대한예측이동시에가능한분류기를설계 Discrimination 분류하기좋게, Intra class 와 Inter class 의분산을딥러닝의학습방식으로최적화 7

연구목표 Structure Deep learning Network Training Learning Feature extraction Learn a Discrimination Testing Correct class label 8

II. 관련연구 Reinforcement Learning and AlphaGo Part.1 9

Feature 제한사항 모노카메라로촬영된이미지영상만으로입력특징을추출 서로다른조건에서동일한사람의큰보행변화는고려하지않음 => 개인의본래보행특성은변하지않는다.[1] Individual Recognition Using Gait Energy Image [2] 새로운시공간보행표현을제시 한장의이미지로, 정적 - 동적인부분을표현 실루엣이미지의노이즈를보완하는 Gait energy image 를제안 10

Feature Gait energy image (GEI) 사람의정기적인걷기는움직임이안정된주파수에서반복되는주기적운동 걸음걸이의전체사이클에서시간정규화된누적에너지이미지 기존대다수의실루엣이미지기반의 Gait recognition 대비 20% 이상의정확도향상 GEI+optical flow, GEI+motion 등의파생된 Feature 알고리즘이있지만, 이러한알고리즘대비간단히 Feature 를생성할수있음, Handcrafted data 의단점을보완 가장많은 Database CASIA, OU-ISIR, USF gait database.. 11

GEI + CNN GEINet: View-Invariant Gait Recognition Using a Convolutional Neural Network [3] GEI 를이용하여, 사람을식별하기위해다량의데이터로 CNN 을처음적용 10,000 명의사람에대한데이터를각도별로직접확보 (OU-ISIR) 2 layer CNN 으로구성 ( 레이어가추가되면성능이저하됨을실험적으로확인 ) 12

GEI + Siamese GEINet: View-Invariant Gait Recognition Using a Convolutional Neural Network [3] 기존머신러닝알고리즘대비, 하여가장성능이좋게측정 (ROC curve, rank) 그러나, 제한적인각도로만성능평가 다른클래스간의차별성을학습하는데한계를보임 13

GEI + Siamese Siamese neural network based gait recognition for human identification [4] 입력간의차별성을학습하기위해, Siamese network 도입 두개의쌍둥이 CNN 을이용하여, 최종출력은두이미지간의학습한거리를출력 가장마지막에 KNN 을사용하여, 거리에따라클래스라벨을부여 + KNN classifier 14

GEI + Siamese Siamese neural network based gait recognition for human identification [4] 두이미지를 pair 하여, 새로운라벨을부여하는전처리작업이필요 모든경우의수에대한 DB 를만들면, 데이터의불균형이발생 랜덤샘플링으로인해, 데이터를충분히활용하지못함 15

GEI + ALL CNN A comprehensive study on cross-view gait based human identification with deep cnns [5] CNN 으로여러가지네트워크를구성하고이를조합하여, fusion 네트워크를설계 모든 DB, 각도, 네트워크구성, 네트워크조합으로모든성능을평가 가장정리가잘된논문 16

GEI + Triplet loss On input/output architectures for convolutional neural network-based cross-view gait recognition [6] 구글의 Triplet loss 방법을, gait recognition 에도입 기존 Siamese 방식의단점을보완하여, 3 input 으로네트워크를구성 Anchor / Positive / Negative 이미지간의유사성, 차별성을각각학습 17

GEI + Triplet loss On input/output architectures for convolutional neural network-based cross-view gait recognition [4] 세가지이미지를 pair 하여, 새로운라벨을부여하는전처리작업이필요 모든경우의수에대한 DB 를만들면, 데이터의불균형이발생 Hard Positive, Nagative 를찾는연산이필요 Unseen data 에취약한성능을보임 18

III. 제안하는방법 Reinforcement Learning and AlphaGo Part.1 19

Reinforcement Learning and AlphaGo Part.1 20

Reference [1] Han, Ju, and Bir Bhanu. "Performance prediction for individual recognition by gait." Pattern Recognition Letters 26.5 (2005): 615-624. [2] Han, Jinguang, and Bir Bhanu. "Individual recognition using gait energy image." IEEE Transactions on Pattern Analysis & Machine Intelligence 2 (2006): 316-322. [3] Shiraga, Kohei, et al. "Geinet: View-invariant gait recognition using a convolutional neural network." Biometrics (ICB), 2016 International Conference on. IEEE, 2016. [4] Zhang, Cheng, et al. "Siamese neural network based gait recognition for human identification." Acoustics, Speech and Signal Processing (ICASSP), 2016 IEEE International Conference on. IEEE, 2016. [5] Wu, Zifeng, et al. "A comprehensive study on cross-view gait based human identification with deep cnns." IEEE Transactions on Pattern Analysis & Machine Intelligence 2 (2017): 209-226. [6] Takemura, Noriko, et al. "On input/output architectures for convolutional neural network-based crossview gait recognition." IEEE Transactions on Circuits and Systems for Video Technology (2017). [7] Wen, Yandong, et al. "A discriminative feature learning approach for deep face recognition." European Conference on Computer Vision. Springer, Cham, 2016. 21

감사합니다 Reinforcement Learning and AlphaGo Part.1 22