Planning is important.
Server Category: Sapeon (X330)
Below is the revised and naturally refined project proposal:
Project Name
AI Table Tennis Referee
Project Topic
By automating table tennis score calculation using AI, we help players focus more on the game.
Project Purpose and Future Applications
In major competitions like the Olympics or World Championships, trained referees watch the game from start to finish and calculate scores. The purpose of this project is to have AI perform the role of these real table tennis referees.
An automatic scoring system helps players focus on the game, and in situations where in/out calls are difficult, it can serve as a supplementary resource for judgment - primarily through the scoring system, and secondarily through replays. Accumulated table tennis match data can be developed into a match history search system or community system for enthusiasts. Beyond table tennis, it can also be applied to other racket sports such as tennis and badminton.
Motivation and Aspirations
I was a table tennis player during high school. After that, I stopped playing competitively and became a developer, but I still enjoy table tennis as a hobby. Every time I played a game, I found it inconvenient to have to remember the score. I had to remember the score for every point and also keep track of who had the serve.
Recently, I encountered VR table tennis games and experienced the convenience of automatic scoring. Through this experience, I thought about how great it would be to have a system that automatically calculates scores in real table tennis games, and I decided to apply because I thought this competition was an opportunity to realize that idea.
During my college years, I had experience with a toy project tracking table tennis balls. Detecting the ball itself was a major challenge. At the time, I succeeded in finding the ball using the Canny edge detection algorithm and Hough transform algorithm through the OpenCV library, but there was a problem where the ball couldn't be detected in environments where the background color matched the ball's color. To overcome this problem, I used table tennis balls of a special color that didn't blend with the background, but in actual games, the ball color is mostly limited to white, so there were limitations in using my algorithm in real games.
I recently learned about SAM2 (Image Segmentation Model), open-sourced by Meta, and saw great potential in the model's ability to accurately identify object boundaries and enable real-time tracking. I thought I could use this model to detect table tennis balls and tables, and build an AI table tennis referee based on that. Furthermore, I believe using products from domestic AI semiconductor companies that rival NVIDIA would make this an even more meaningful project. Through this competition, I want to develop software that creates a wonderful experience for 300 million table tennis players worldwide by building an AI table tennis referee.
AI Algorithms and Technologies to Be Used
Match footage is transmitted in real-time to the server through cameras installed at the table tennis table. The positions of the table tennis ball and table are detected from this footage using SAM2 (Segmentation Anything Model). Based on table tennis game rules, scores are calculated using the detected positions of the table and ball, along with the computed trajectory of the ball.
1. Computer Vision
Match footage is transmitted in real-time to the server through high-resolution cameras installed at the table tennis table. Based on this footage, the positions of the table tennis ball and table are accurately determined.
- Object Detection: For accurate detection of the table tennis ball and table, we utilize SAM2 (Segmentation Anything Model) released by Meta. This model precisely distinguishes the boundaries of the ball and table, tracking the ball's position and trajectory in real-time.
2. AI Model Inference Service Using Server-Based NPU
-
Streaming Data Processing: To process real-time match footage, video data is reliably transmitted and processed through socket communication programs. Additionally, edge computing technology is used to perform initial data preprocessing and filtering on edge servers near the venue, with only key data transmitted to the central server.
-
API and Service Deployment: Based on the optimized model, real-time scoring and judgment services are provided through RESTful APIs. The API server is containerized and deployed using Docker and Kubernetes.
3. Scoring System Integration
The detected table tennis ball's position and trajectory information is transmitted to a scoring system that calculates points based on table tennis game rules.
-
Automatic Score Calculation: The ball's movement trajectory is analyzed to determine in/out calls, and scores are calculated in real-time. Especially in ambiguous situations, automatic judgment is supplemented with a video replay function to assist decisions.
-
Serve Tracking: Following table tennis game rules, the serve automatically switches every 2 points, and serve positions and serve changes during the match are recorded and communicated to users.
-
Data Storage and Analysis: All data generated during the match is stored in a database for post-match analysis.
SLLM Category: Rebellions ()
AI Semiconductor Tech Talent Selection Competition Project Proposal
Name (Kim Youngju)
Project Name
Project Topic
Project Purpose and Expected Future Applications
Motivation and Aspirations
AI Algorithms and Technologies to Be Used
https://www.aichipcon.or.kr/apply/applyInfo
Things I'm subscribed to, things I need to do.
- Docker Desktop
- Netflix
- Youtube
- KodeKloud
- Colab
- Chat GPT
I want to grow. This is a time when I need to grow. Let's not settle for the sweet and comfortable present. Words seem to be a really important factor.
Let's study Git Ops. Argo CD. Backend, Front end, Infra (Network, OS), Cloud (AWS), DevOps.
I want to store Text Data in Elastic Search (Open Search) and try searching through it.