WELCOME
About me
Research interest: Computer vision
- Re-identification, Object tracking
- Transfer learning, Pose transfer
- Multi-modal image retrieval
- Deep learning on robotics
- Gesture recognition
Latest News
- Be awarded the Asian Federation Computer Vision(AFCV) best paper award on kroc2020 (2020.08.17)
- Presented ''Pose transferred Image Generation Model Applicable to IR Images" on kroc2020 (2020.08.15)
- Presented "RGB-IR person re-identification with pose transferred image generation" for defense (2020.06.13)
Educational background
Goyang Foreign Language High School (고양외고)
2011. 03 ~ 2014. 02
Chinese Language
Ulsan National Institute of Science and Technology (UNIST - 유니스트)
2014.03 ~ 2018.08
Mechanical and Aerospace Engineering - Major, Human Factors Engineering - 2nd Major (B.S)
Korea Advanced Institute of Science and Technology (KAIST - 카이스트)
2018.08 ~ 2020.08
Mechanical Engineering (M.S) - Major
Deep Learning based Image Retrieval - Major Research Field
Research
Keywords: Computer vision, Deep learning, Re-identification , Multi-Modality data, Generative Adversarial Network(GAN), Image generation, Touch gesture recognition, 1-D Convolutional Neural Network (1D-CNN), Pose transfer, Infrared(IR) Image
RGB-IR Cross-Modality Person Re-Identification with Pose-transferred Image Generation
Main Research - Thesis
Data
- RGB image, IR image
* SYSU-MM01 Public data
Method : Deep learning
- Siamese network
- Generative adversarial network
--> Pose transferred image generation
--> RGB to IR image generation
Contribution
- The first attempt of introducing pose transfer in RGB-IR Re-ID
- Validation of pose transfer in all types of different pose variance gallery set
- The greatest performance increment in the high pose variance gallery set
- Showing that singe generator-based approaches produce better performance in robotic application
Data
- IR image
Method : Deep learning
- Generative adversarial network
--> Pose transfer
Contribution
- Propose the IR pose transfer which can be applicable to infrared images
- Modification in the generation method of the heat map
- Alleviation of blurred phenomenon (qualitative evaluation)
- Numerical improvement on all 5 similarity evaluation scores (quantitative evaluation)
Touch Gesture Recognition System based on 1D CNN with Two Touch Sensor Orientation Settings
Data
- Visualized gray image
Method : Deep learning
- 1-D Convolutional Neural Network
Contribution
- Developed a touch gesture recognition system that can distinguish four touch gestures: hit, pat, push, and rub
- Showing 90.5% average recognition accuracy, which is 29.4% higher than the that of the related work based on TDT
- Confirming the effect of touch sensor orientation on recognition performance
- Newly constructed system showed high recognition performance on both vertical and horizontal datasets
Research Award
Asian Federation Computer Vision(AFCV) best paper award in the 15th Korea Robotic Society annual Conference (KROC 2020)
Research Publication (First author)
J.-H. Park, J.H. Seo, Young-Hoon Nho, and D.-S. Kwon, Touch Gesture Recognition System based on 1D Convolutional Neural Network with Two Touch Sensor Orientation Settings, 2019 16th International Conference on Ubiquitous Robots (UR) (pp. 65-70). IEEE.
J.-H. Park, J.H. Seo, and D.-S. Kwon, Pose transferred Image Generation Model Applicable to IR Images, The 15th Korea Robotics Society Annual Conference (KROS)
Project
1. 산업 및 복합생활공간 생활안전 AI 서비스 검증을 위한 리빙랩 구축 및 운영(2018 ~2020) - 한국건설생활환경시험연구원
2. 지능형 로보틱스 기초 기술 연구(2020) - KAIST
3. 사용자 맞춤형 감성 교감 기술 개발(2018) - LG전자
1. 산업 및 복합생활공간 생활안전 AI 서비스 검증을 위한 리빙랩 구축 및 운영(2020년도)
Development of an intelligent platform for living safety service in complex living space
Project description
- 주거 환경 속 영유가 화상 사고 예방 시스템을 구축하기 위해 인공지능 기반의 위험/비위험 상황 감지 시스템 개발
Detailed description
- 리빙랩 구축 및 영유아/보호자 분류기 학습을 위한 사용자 일상 생활 데이터 수집 (RGB image, Thermal image)
- 사람 Detection을 위해 YOLO (대표적인 물체 감지 기술) 알고리즘 활용
- 인식기의 신뢰도 및 안전성을 높이기 위해 YOLO와 SORT(대표적인 물체 트레킹 기술) 연동
- 영유아/보호자 분류기 모델의 소형화 및 Low Computational power PC에 적용
2. 지능형 로보틱스 기초 기술 연구(2020) - KAIST
Intelligent robotics technology research
GAN-based proposed approaches
Project description
- 서비스 로봇의 인간과의 자연스러운 상호작용 위한 기반 요소 기술인 감정인식기와 사용자 신원 파악 기술 개발
- 사용자 신원 파악 기술 중 조명과 포즈에 강인한 기술 개발
Detailed description
- RGB-RGB 사람 재식별 기술 중 사람 포즈 변화의 강인한 FD-GAN를 밴치마킹
- FD-GAN을 기반으로 포즈 변화에 강인한 RGB-IR 교차 모달리티 사람 재 식별 기술 개발
- 두가지 접근 방식인 단일 생성기 접근 방식과 멀티 생성기 접근 방식 기반으로 개발
- 성능 평가를 통해 RGB to IR modality translation기능과 Pose transfer 기능을 한 인코더에 - 학습한 단일 생성기 접근 방식 모델이 로봇 적용에 있어 더 유효함을 확인
3. 사용자 맞춤형 감성 교감 기술 개발(2018)
Development of customized emotional communication technology
Data visualization of touch gesture
The overall structure of the touch gesture recognition system
Project description
- 서비스 로봇과의 감성적인 교감을 위해 필요한 기술 요소로 인공지능 기반의 터치 제스터 인식기 개발
Detailed description
- 대표적인 터치 제스쳐 4가지의 데이터 수집 및 이미지 시각화 진행
- 위 이미지의 흰 색 영역이 Touched Cell, 검은 색 영역이 Non-touched Cell을 의미
- 이미지로 visualization 시켰을때 각 touch gesture 의 패턴이 분명하여, 이를 처리하는데 좋은 성능을 보이는 Convolutional Neural Network을 활용
- Computational Load를 줄이기 위해 1-Dimensional CNN 이용
RESEARCH ACTIVITIES
ACADEMICAL ACTIVITIES / PROJECT
Vision-related academic courses during M.S : Visual Intelligence, Human-Robot Interaction, 4th Industrial Revolution & Innovation, Pattern Recognition, Machine Learning for Knowledge Service, Advanced Deep Learning
Development of facial expression recognition
감정 인식기 개발
Development of handwriting based character and word recognition
필기체기반 글자 및 단어 인식기 개발
OTHER ACTIVITIES
2014. 02. – 2016. 02. : Planning Director of Student Ambassador in UNIST
2017. 03. – 2017. 08. : Research Intern at Ergonomics Laboratory in UNIST
2018. 09. – 2019. 02. : Special Lecture in Mechanical Engineering Course Academic Assistant in KAIST
2019. 09. – 2020. 08. : Mechanical Engineering Counseling Assistant in KAIST
Activities & Skill
KROC Conference
한국로봇학회 발표
2020.08.17
Master Thesis Defense
학위연구발표
2020.06.13
URAI Conference(IEEE)
URAI 학회발표
2019.06.24
TECHNICAL SKILL/EXPERIENCE
Operating System(OS): Ubuntu, Window, Robot operating system(ROS)
Programming language: Python, MATLAB, c/c++, Shell scripting
Machine learning opensource platform/library: Pytorch, Tensorflow
GUI programming: qt creator
Other programs: Pycharm, Catia, Labview, Ansys..
CERTIFICATE
IoT knowledge test certificate (IoT 지식능력검정시험)
LANGUAGE PROFICIENCY
Basic comprehension ability in Chinese - New HSK 5 (2017.05)
Fluent in English - Toeic speaking level 6 (2020.08)
Native in Korean
MOTTO
Life quotes
Be the reason someone believes in the goodness of people
스스로 누군가가 사람의 선량함을 믿게 되는 이유가 되길..
Karen Salmansohn
캐런 살만손
How To Contact
서울특별시 용산구 한남대로
Hannam-daero, Yongsan-gu, Seoul, Korea
+82 10 8745 0720