Oh-Hun Kwon — AI / Computer Vision / Robotics

About Me

I am a Computer Vision and Robotics Engineer with a strong focus on 3D perception, multi-view systems, and real-world AI pipelines.

My work combines geometric reasoning, deep learning, and system-level engineering to build scalable and efficient solutions. I have extensive experience designing multi-camera systems, calibration pipelines, and GPU-accelerated processing using PyTorch and CUDA.

I enjoy working on problems that connect perception with real-world systems, such as tracking, reconstruction, and sensor fusion.

Experience

Freelance AI / Computer Vision Engineer

Jan 2025 – Present

Consulting and implementation for computer vision and 3D perception systems
Designed and prototyped vision pipelines (data processing, evaluation, integration)
Developed LLM-based retrieval systems focused on data organization and system integration

Research Engineer — University of Bonn

Sep 2020 – Dec 2024

Built multi-camera systems for 4D plant phenotyping.

System 1 – RGBD Prototype

Developed multi-view capture system using RGBD cameras
Implemented stereo matching to improve depth accuracy
Automated acquisition pipeline and hardware control

System 2 – High-Resolution Multi-Camera System

Designed rotating multi-camera capture system
Built synchronization and control pipeline
Developed a novel PyTorch-based calibration method using turntable kinematics

Research Assistant — University of Bonn

Apr 2018 – Aug 2020

Contributed to the Bonn Activity Maps dataset
Managed multi-camera data acquisition (12 cameras)
Built synchronization and preprocessing pipelines
Resolved critical calibration issues with a custom 3D keypoint localization method

Teaching Assistant — Computer Vision 1

Oct 2017 – Mar 2018

Researcher — KAIST

Sep 2008 – Feb 2015

Worked on multiple robotics systems:

7-DOF motion simulator (control system design)
Smart rehabilitation walker
Autonomous driving behavior analysis
Mobile robot modeling
Telepresence robot for remote education

Selected Projects

🚀 OctTorch — GPU-Accelerated Octree Volume Carving

High-performance 3D reconstruction using PyTorch + CUDA

Scales to gigavoxel resolution
Custom GPU data structures and kernels

ohkwon718/OctTorch

📷 cali360 — Multi-Camera Calibration Toolkit

Calibration pipeline for turntable-based camera systems

Animated preview of cali360 multi-camera calibration and visualization

PyTorch-based optimization
ArUco-based calibration
Interactive visualization

ohkwon718/cali360

🎯 Bayesian Multi-View Pose Tracking

Multi-view Bayesian pose tracking preview

Recursive Bayesian filtering for multi-camera tracking
Combines deep learning detections with probabilistic estimation
Published at ACCV 2020

ohkwon718/bayesian-mvpose

🌐 Gaussian Splatting (Educational Implementation)

Gaussian splatting demo view from toy_GS

Minimal implementation for neural rendering
Integrated into Open3D GUI for experimentation

ohkwon718/toy_GS

🧱 Open3D App Template

Framework for building interactive 3D applications
Clean architecture (UI / Controller / Domain separation)

ohkwon718/open3d-app-template

🎬 Video Synchronization Tool

Multi-camera synchronization using audio FFT

ohkwon718/multi-mp4-sync

📐 Camera Calibration Tools

Multi-camera calibration pipeline preview

Intrinsic & extrinsic calibration pipelines
Optimization-based parameter estimation

multicam-calibration calib-with-mp4

📊 Distributed Kalman Filter

Implementation and visualization of distributed filtering

ohkwon718/Distributed_Kalman_Filter

Publications

2023

Tanke, J., Kwon, O.-H., Mueller, F. B., Doering, A., & Gall, J. (2023). Humans in kitchens: A dataset for multi-person human motion forecasting with scene context. Advances in Neural Information Processing Systems (NeurIPS), 36, 10184–10196.
Kwon, O.-H., & Zell, E. (2023). Image-coupled volume propagation for stereo matching. In Proceedings of the IEEE International Conference on Image Processing (ICIP). · Code

2020

Kwon, O.-H., Tanke, J., & Gall, J. (2020). Recursive Bayesian filtering for multiple human pose tracking from multiple cameras. In Asian Conference on Computer Vision (ACCV). · Code

2019

Tanke, J., Kwon, O.-H., Stotko, P., Rosu, R. A., Weinmann, M., Errami, H., Behnke, S., Bennewitz, M., Klein, R., Weber, A., Yao, A., & Gall, J. (2019). Bonn Activity Maps: Dataset description. arXiv preprint arXiv:1912.06354.

Skills

Computer Vision: Camera Calibration, Epipolar Geometry, Optical Flow, Segmentation, Object Tracking, 3D Reconstruction
Machine Learning: Bayesian Filtering, SVM, GMM, HMM, Clustering, Reinforcement Learning
Deep Learning: CNN, RNN, Transformers, Diffusion Models
3D & GPU: PyTorch (advanced tensor ops), CUDA, Open3D, Gaussian Splatting (3D)
Programming: Python, C++, OpenCV, NumPy, Eigen
Robotics: ROS, SLAM, Kinematics, Inverse Kinematics, Motion Planning, Control Systems
Systems & Tools: Linux, Docker, Git, CMake, Bash, Unreal Engine (C++, Blueprints), Blender (3D modeling)
LLM & AI Systems: RAG, LangChain, vLLM, Agent-based systems
Cloud: AWS, Azure
Languages: Korean (Native), English (Fluent), German (Basic)

Education

M.Sc. Computer Science — University of Bonn

2016 – 2020

Thesis: Multi-Person 3D Pose Tracking

M.Sc. Robotics — KAIST

2008 – 2010

B.Sc. Electrical Engineering & Computer Science

Kyungpook National University