Projects
GIST-MIT, Collaborative Project on Contrastive Learning and Resource Limited Devices (August 2022~ Present)
In this collaborative project with the Massachusetts Institute of Technology CSAIL. I have been working on federated learning, On-device machine learning, and contrastive learning research to address the generalization and non-idd problems.
Abstract --> Due to the open-set nature of face recognition, face recognition models usually meet the different unseen domains and environments in real-world applications. A well-trained face recognition model shows unsatisfactory performance due to different illumination conditions and motion blur between registered candidates and target candidates. More specifically, a model trained on a central dataset shows poor generalization between the unseen same candidates with different illumination and blur conditions. In this paper, our goal is to learn better generalization for different target conditions without any active or incremental learning. We propose a novel face recognition training approach by using contrasting learning to address the generalization problems in face recognition models.....
AI-based Face Detection, Face Recognition on Infrared Thermal camera Feb~2020
During the CORONA pandemic, it was very essential to monitor the individual's body temperature with their identity. With the help
of machine learning algorithms and fusion technology, I have developed a thermal and RGB camera algorithm to recognize identity and monitor an individual's body temperature.
This project was developed using sensor fusion techniques. First, the thermal and RGB camera frames are collected to calibrate the two cameras. The infrared heater is used to heat up the calibration check board and then it has been shown to the thermal camera to detect the chess corner points. The deep learning model is used to simultaneously detect and recognize faces in an RGB camera. The detections are mapped in a thermal camera using a translation matrix by multiplying the camera coordinates with the calibration matrix. The absolute temperature is selected from the face to identify individual body temperature. The project was successfully installed in the EECS, GIST department at the beginning of Feb 2020 when temperature monitoring for necessary during the COVID pandemic.
This project was covered by local news https://www.dongascience.com/news.php?idx=37810
Development of Face Recognition Device & Startup Company June 2020~ Present
This project includes the development of a tiny machine learning model for IoT edgetpu devices to accurately perform adaptive wild face recognition. This project was designed solely and the roles are to design a 3D model, electronic hardware (including PCB and SoM), and software framework implementation. The research further extended to federated learning in IoT devices in collaboration with the GIST-MIT project. This project was initially started as a product for the startup company with my advisor Prof. Moongu Jeon
Project Page smart-face.github.io/smartface/
Development of Golf Cart Autonomous Vehicle Project
This project is about designing hardware and software for golf carts, which include automatic steering, brake, and throttle controls. For that purpose I have designed own steering angle sensor to be able feed back and control the steering from the software.
The main objective is to enable software-based control for the golf cart which can be controlled by perception algorithms and planning algorithms.
Development of Multi-object Tracking Algorithm - Deep View ICT Project
Designing a multi-object tracking algorithm for the ICT project.
