TRACY

Logo and Thumbnail
Landing page
Personalized AI analysis and coaching
Ball trajectory and bounce data
Game analysis GUI (heatmaps, tracking data)

Inspiration

Our inspiration for TRACY came from the desire to enhance tennis training through advanced technology. One of our members was a former tennis enthusiast who has always strived to refine their skills. They soon realized that the post-game analysis process took too much time in their busy schedule. We aimed to create a system that not only analyzes gameplay but also provides personalized insights for players to improve their skills.

What it does and how we built it

TRACY utilizes computer vision algorithms and pre-trained neural networks to analyze tennis footage, tracking player movements, and ball trajectories. The system then employs ChatGPT for AI-driven insights, generating personalized natural language summaries highlighting players' strengths and weaknesses. The output includes dynamic visuals and statistical data using React.js, offering a comprehensive overview and further insights into the player's performance.

Challenges we ran into

Developing a seamless integration between computer vision, ChatGPT, and real-time video analysis posed several challenges. Ensuring accuracy in 2D ball tracking from a singular camera angle, optimizing processing speed, and fine-tuning the algorithm for accurate tracking were a key hurdle we overcame during the development process. The depth of the ball became a challenge as we were limited to one camera angle but we were able to tackle it by using machine learning techniques.

Accomplishments that we're proud of

We are proud to have successfully created TRACY, a system that brings together state-of-the-art technologies to provide valuable insights to tennis players. Achieving a balance between accuracy, speed, and interpretability was a significant accomplishment for our team.

What we learned

Through the development of TRACY, we gained valuable insights into the complexities of integrating computer vision with natural language processing. We also enhanced our understanding of the challenges involved in real-time analysis of sports footage and the importance of providing actionable insights to users.

What's next for TRACY

Looking ahead, we plan to further refine TRACY by incorporating user feedback and expanding the range of insights it can offer. Additionally, we aim to explore potential collaborations with tennis coaches and players to tailor the system to meet the diverse needs of the tennis community.

Built With

ffmpeg
keras
matplotlib
opencv
python
react
tensorflow

Submitted to

QHacks 2024
- Winner 3rd Prize

Created by

we do a bit of back end shenanigans here

Daniel Lu
o/
On the backend, I contributed to utilizing ffmpeg to extract only the important clips from an uncut video. I also did most of the work in tracking players/ball/court.

On the frontend, I designed the logo, made the loading screen, and created the video, entirely from scratch :))))) I was also responsible for the video viewer component in the GUI.

Vu Cao
pls enjoy game
On the frontend, I worked on the creation of the web app structure and leading the organizational space. I enhanced the user experience by building the interactable components.

On the backend, I worked on the programs for generating the heat maps for the player and ball. As well, I had worked on the program for the ball tracking trajectory graphs.

Edwin Ngui