Industry-collaboration project with Fitogether focused on integrating GPS and video trajectory data for formation analysis.

Contributed to model validation using Docker and CVAT to verify bounding boxes and labeling accuracy. 

Built a full pipeline to visualize real-time formation changes in soccer broadcast videos.

Utilized YOLO and an object detection model trained in Roboflow to detect and track players.

Applied homography transformation techniques to map player positions from video frames onto a 2D minimap for tactical analysis and visualization. 

Developed a baseball game simulation using pitch-by-pitch data in a team project with Yonsei Sports Analytics Lab and Sports2i Corporation.

Leveraged dynamic data loading, event-based game logic, and self-developed scoring algorithms with Pandas and Matplotlib

A team project at Yonsei Sports Analytics Lab, in collaboration with Sports2i, to develop a volleyball block touch-out detection system.

Utilized OpenCV, YOLO, MediaPipe, and Roboflow for analyzing player–ball interactions in video footage.

Designed a custom touch-out criterion using correlation coefficients in Pandas to improve detection accuracy and consistency.

Implemented bounding box occlusion checks and Euclidean distance–based hand tracking for robust touch event identification. 

Team project at Yonsei Sports Analytics Lab applying machine learning to classify MLB players and analyze cluster patterns.

Processed data, applied dimensionality reduction (UMAP, PCA, t-SNE), and clustered players using K-means, KMedoids, and DBSCAN.

Optimized with KElbowVisualizer, validated using silhouette scores, identified representative players, and visualized profiles with Seaborn and Matplotlib. 

Personal project inspired by UC Berkeley Sports Analytics Group Berkeley.

Built a linear regression model to predict NBA players’ points per game using 2023 season data.

Analyzed features like minutes, field goals, and assists, achieving MAE 2.46, MSE 10.24, RMSE 3.20 with Python and historical NBA data (1947–2024) from Kaggle.

Developed a web crawling tool using BeautifulSoup to generate ranked lists of trending news topics from the news platform “Nate”.

Extracted and filtered website features to display the top 5 news articles based on user-selected dates.

Awarded the Gold Prize at the 2022 KYIC Korea Invention Idea Contest and presented at the IIIC Conference.