지능형영상보안과 휴먼인식기술 유 장희 ( 모바일바이오인식신융합기술표준연구회, 2012.05.31)
목차 1. 기술현황 2. 지능형영상보안 3. 그미래 : 휴먼인식 4. ETRI 휴먼인식기술 5. 결론 2
1. 기술현황 3
영상보안기술 (1/2) Models, techniques and systems for acquiring videos about the 3D external world, detecting targets along the time and the space, recognizing interesting or dangerous situations, generating real-time alarms recording meaningful data about the controlled scene. Intra-internet Data Acquisition Knowledge Extraction DBMS Surveillance Output Production Human/machine based Control Center 4
영상보안기술 (2/2) Research World Commercial World 60-70 Hardware 60-70 Analogue cameras 80 Military research 80 Digital CCTV systems 90 Traffic monitoring 00 People surveillance 10 Life surveillance 90 Digital surveillance systems 00 Network surveillance systems 10 Intelligent surveillance systems 5
영상보안기술의진화 (1/2) Networked Cameras PC independent - has own computer Web server - email client - FTP client & server Plugs into any Ethernet RJ-45 jack - IP addressable Programmable with many advanced features 6
영상보안기술의진화 (2/2) Intelligent Video Analytics (IVA) Runs on Windows - Unix - Linux servers Monitor cameras - control recording and image analysis Advanced capabilities available Multiple recording modes PTZ control Alarm management and motion detection Full duplex audio 7
CCTV 카메라설치규모 전세계 3,000만대이상의 CCTV 카메라설치 (MIT 등 ) 영국 ( 인구대비 CCTV 수가가장높은나라 : 14명당 1대 ) 440만대이상의 CCTV 카메라설치추산 런던의경우평균 30초에한번씩노출 CCTV 설치후 30% 범죄감소효과분석 미국 (CCTV 가가장많은나라 : 20 명당 1 대 ) 1천500만대추산, 9.11 테러이후급격히증가 국토보안, 재난방재등공공수요의비중이높음 한국 ( 세계적인 CCTV/DVR 장비수출국가 ) 274만대이상의 CCTV 설치추산 (18명당 1대 ) 서울시민, 하루 140회노출 교통흐름및법규단속, 스쿨존, 외곽경계 방범용 (360도회전가능, 12배이상의줌인 ) 8
전세계영상보안시장규모 영상보안시장은연간 11.8% 의성장이예측되며, 지능형영상보안소프트웨어및솔루션은연간 30% 의고성장이예측됨 세계시장의지역별규모로는미주 (48%), 유럽 (20%), 아시아 (17%), 중남미 (6%) 등의순으로형성 12 ( 단위 : Billion US$) 10.22 10 8.51 8 6 4 2 5.89 7.09 0 2007 2008 2009 2010 ( 출처 : 영상보안시장전망, J. P. Freeman, 2007) ( 출처 : 지능형영상보안시장전망, J. P. Freeman, 2006) 9
전세계영상보안시장성장률 품목 2006 2007 2008 2009 2010 2011 CAGR 아날로그장비 3,703 3,870 4,054 4,188 4,271 4,335 3.20% DVRs 2,170 2,423 2,670 2,893 3,095 3,283 8.60% NVRs 59 92 132 176 223 273 35.80% IP 카메라 372 536 763 1,076 1,509 2,056 40.80% 감시 / 관리 SW 158 220 299 407 548 728 35.70% 비디오서버 228 314 429 579 764 998 34.40% 합계 6,690 7,455 8,347 9,319 10,410 11,673 11.80% ( 출처 : IMS Research 2008, 삼성 SDS: 융합보안시장동향보고, 2010/09) 단위 : 백만 US 달러 10
주요이슈 (1/2) Resolution (CIF, D1, Mega) Connectivity (Wired, Wireless, PoE, PLC) Multi-Codec (Mpeg4, MJpeg, H.264) Multi-Profile (Cropping, Digital-PTZ) Intelligence (Motion tracking, Fire Detection, ) Night Vision (Enhanced IR, TDN, WDR, ) Distributed Storage (Camera, DVR, NVR, ) Mobile (3G, ) Standardization (PSIA, OnViF) 11
주요이슈 (2/2) Resolution Revolution All about pixels per distance By 2012, 50% of all new camera will be high resolution camera Decentralization Camera itself is the DVR Moving the DVR functionality from a central monitoring system to the camera Automatic Video Analysis Video-based detection and tracking Video-based person identification 12
2. 지능형영상보안 13
지능형영상보안기술 Definition: Data captured by video surveillance cameras Real time and automated analysis of video sequences Video understanding= from people detection and tracking to behavior recognition Recognition of complex behaviors: of individuals (e.g. fraud, graffiti, vandalism, bank attack) of small groups (e.g. fighting) of crowds (e.g. overcrowding) interactions of people and vehicles DARPA Projects: Video Surveillance and Monitoring (VSAM) Human ID at a distance (HID) Combat Zones That See (CTS) 14
지능형영상보안요구 국내 CCTV 설치현황 2010년현재 274만대이상 감시인력부족 ( 주로사후대응용 ) 스쿨존 : 9,609 곳지정 2010년 706억원예산투입 4,225대추가설치 감시요원의주의력지속시간 ( 모니터 2 대기준, 미국법무성 ) 15
People, Traffic, and Environment People: Traffic: Security - People detection in specific area - People tracking for abnormal behavior detection - People interaction analysis Safety - People behavior monitoring in house and offices - People behavior analysis in public place or car Road traffic control: statistical control and accident detection Urban control: people at road crossing, privileged path, ) Naval/Aerial traffic control Environment: Natural environment (fire detection, fog detection, ) Urban environment (crowd monitoring in stations, subway, ) Abnormal situation detection 16
지능형영상보안모델 Video Change Detection Region Localization Region Tracking Models User Object Classification Queries Object Identification Movement Classification Object Tracking Models Activity Classification 17
지능형영상보안시스템분석 ObjectVideo, 3VR, IBM ( 미국 ), iomniscient ( 호주 ), ioimage ( 이스라엘 ), Bosch ( 독일 ) 가대표적이며, 이들의시장점유율이가장높음 : 학계연구 : 업체제품 Change Detection Region Localization Region Tracking Object Classification Object Tracking Object Identification Movement Classification Activity Classification VSAM/CMU Sarnoff VSAM/MIT Remagnio et al. Aimetis BOSCH IVA Honeywell IBM SSS iomniscient ioimage ObjectVideo 18
3VR IVMS (1/2) 19
3VR IVMS (2/2) 얼굴인식 문자인식 상태감시 20
ObjectVideo and Bosch IVA ObjectVideo OnBoard 응용시스템 Bosch 제품을이용한응용시스템구성 21
iomniscient VMS 22
Video Surveillance Challenge PETS IEEE Workshop on Performance Evaluation of Tracking and Surveillance (PETS) 11 th Workshop, held in conjunction with CVPR 2009 AVSS (i-lids and SPEVI) IEEE Conference on Advanced Video and Signal based Surveillance (AVSS) Surveillance Performance EValuation Initiative (SPEVI) CLEAR CLassification of Events, Activities and Relationships Evaluation and Workshop 23
주요이슈 Systems have poor performances over time, can be hardly modified and do not use enough a priori knowledge tiny objects lighting conditions strong perspective close view shadows clutter 24
3. 그미래 : 휴먼인식 25
휴먼식별 : Biometrics 26
휴먼식별 : Human ID at a Distance 27
원거리휴먼식별 : HID Project (1/2) Cooperative Subjects Single Mode Biometric Limited Distance (< 10 Feet) Identification/Verification Simple Systems Applications Face at a Distance Non-cooperative Subjects Multi-mode Measurements Distance to 500 Feet Person in a Crowd Identification Science Multiple modes at a distance 28
원거리휴먼식별 : HID Project (2/2) Face Recognition Gait Recognition Iris Recognition Psycho-Physics Advanced Sensors System Development 29
원거리홍채인식기술 IRIS on the Move ETRI BioGate AOptix InSight ( 출처 : www.sarnoff.com/iom) ( 출처 : www.aoptix.com/) 30
원거리얼굴검출및인식 31
원거리휴먼인식 : Gait Emerging gait research Gait is non-contact and uses sequences Advantages: perceivable at distance and hard to disguise Potential applications: security/ surveillance, immigration, forensic, medicine? Other applications: moving objects Related fields: animation, tracking, robotics 32
4. ETRI 휴먼인식기술 33
ETRI 바이오인식솔루션 (1/2) 얼굴 / 홍채 / 지문인식 SDK Windows, WinCE, Linux, embedded Linux 지원 100 개이상의관련특허확보 지문센서 지문영상획득모듈 지문영상 지문특징추출모듈 1 지문영상캡춰 영상전처리 a 특징점추출경계특징점제거 지문템플릿 2 저장장치 c 지문특징정합모듈지문영상보정 b d 등록과정매칭결과매칭 Y/N 인증과정 Fingerprint SDK & System Iris SDK and System Face SDK & System 34
ETRI 바이오인식솔루션 (2/2) 응용시제품 BioGate ( 로비폰, 도어폰, Kiosk) 모바일 /DVR용얼굴인식솔루션 원거리홍채인식및얼굴검색시스템 Iris-Fingerprint Door phone Face-Fingerprint Lobby Phone Biometric Kiosk for e-passport Applications Face-Fingerprint/Iris- Fingerprint Kiosk Mobile Device for Face Identification Face Retrieval System for Large Database Face recognition for DVR System Iris recognition at a Distance 35
주요기술현황 (1/4) 지문보호형 e-id Kiosk One-Time Fuzzy Vault 일회성템플릿을이용하여인증시마다템플릿을갱신 상관관계공격등과같은공격을원천적으로차단 지문특징점이유출되더라도유출된특징점을재사용할수없도록함 10 만개이상의가짜특징점삽입가능 FMR 이 0 ( 은행, 전자주민증응용에강점 템플릿크기경량화 소수의렌즈 / 카메라사용으로인한가격경쟁력우위 e-id 카드환경에서구동가능한템플릿경량화 Geometric Hashing 기법 : 2~3Mbytes 전수조사에의한자동정렬적용 : 25KB 이하 퍼지볼트기반지문정보보호모듈 36
주요기술현황 (2/4) 사용자친화형홍채인식기술 원거리홍채인식카메라 70~100cm 거리에서홍채인식 750nm~850nm 적외선조명 홍채지름이 150 픽셀이상의눈영상획득 (15fps 이상 ) 사용자홍채입력상태확인용 LCD 모니터 눈영역탐지눈영역추출홍채인식 SDK 홍채영역추출 홍채 / 얼굴인식다중카메라 특징추출및코드생성 Beam Splitter/Dichroic Mirror 사용 ( 홍채 / 얼굴영상동시획득 ) 홍채코드매칭 원거리홍채인식과얼굴인식결합가능 소수의렌즈 / 카메라사용으로인한가격경쟁력우위 주요특징 적외선홍채영상과가시광선얼굴영상동시획득가능 객체감지후카메라 On-Off 얼굴영역탐지 얼굴영역추출 얼굴인식 SDK 특징추출및얼굴템플릿생성 얼굴템플릿매칭 Iris Database Face Database 국내특허및국제특허출원 37
주요기술현황 (3/4) 얼굴검색시스템 주요특징 비대면얼굴인식기술 포즈, 조명, 가림, 저해상도등에의한얼굴인식성능저하극복 범죄수사 / 영상보안등에활용가능한얼굴검색 데이터베이스 얼굴인식서버 XML Access Point 핵심기술 얼굴검색요청 얼굴검색결과 Mobile Device 포즈 / 조명에강인한얼굴인식 얼굴복원기술 ( 초해상도얼굴복원, 가림얼굴복원 ) 대용량얼굴검색기술 활용분야 범죄자검색등첨단수사분야 출입국관리분야 미아찾기등유사얼굴검색서비스분야 지능형 CCTV 영상보안분야 38
주요기술현황 (4/4) 원거리사람식별및검색 주요특징 객체정보획득, 검출, 추적및분류전자태그연동원거리사람식별 CCTV 연동원거리사람검색사람추적및비정상행동패턴인식 핵심기술 스마트카메라및다중플랫폼지원사용자친화형사람인증대용량 DB 검색포즈 / 조명에강인한사람인식 활용분야 사용자친화형출입통제범죄수사, 위험인물색출 CCTV, 스마트카메라출입국관리분야지능형 CCTV 영상보안분야 39
원거리휴먼인식 : 시스템구조 원거리사람검출및추적 주요특징 고정 / 이동카메라기반사람 / 객체검출고정 / 이동카메라기반사람 / 객체추적다중카메라연동기반객체검출 / 분리원거리얼굴검출및검색서비스 핵심기술 고정카메라사람검출률 90% 이상이동카메라사람검출률 90% 이상실시간배경과객체분리율 85% 이상고정-PTZ 카메라객체검출 90% 이상 활용분야 지능형 CCTV 영상보안분야사용자친화형출입통제요주의인물검색서비스지능형로봇인공시각시스템지능형자동차전장안전출입국관리분야 40
원거리휴먼인식 : 비대면얼굴검색 얼굴검색시스템 - 자동등록 / 조회 모바일얼굴검색 - 모바일단말연동 CCTV 얼굴검색 - 고정 -PTZ 연동 41
원거리휴먼인식 : 실시간얼굴검출 42
원거리휴먼인식 : 실시간얼굴검색 43
원거리휴먼인식 : 모바일얼굴검색 44
5. 결론 45
요약및결론 단순모니터링위주의노동집약적감시환경으로부터지능적 / 자율적감시환경의구축을위한영상분석솔루션에대한요구는지속적으로증가될전망됨 영상보안솔루션공급업체의 90% 이상이향후지능형솔루션의제공계획을가지고있음 그리고이러한새로운요구는현재의영상보안기술이도달해야할그리고극복해야할최종적인목표가될것임 현재다양하고화려한기술들이존재함에도불구하고사용자예상과이들시스템의성능에커다란차이가존재함 향후영상보안기술은원거리휴먼식별등과의결합을통한보다다양하고새로운서비스의개발이진행될것임 46
연락처 : 휴먼인식기술연구팀 유장희팀장 (042-860-1324, jhy@etri.re.kr) 감사합니다. Q&A