-->
We assisted an AI startup in stealth mode to build its core platform for analyzing live or recorded squash games. This project aimed to enhance squash court analytics through innovative technological interventions, addressing the challenges of accurately tracking player movements and ball trajectories.
Our client is a startup currently in stealth mode, dedicated to developing an application for squash players and coaches. They sought technical assistance to build their core product, which needed to analyze live or recorded squash games effectively.
Accurately track player movements and ball trajectories.
Overcome the limitations of existing analytics methods, which lacked precision and efficiency, hindering comprehensive insights into player performance and game dynamics.
Utilized perspective transformation techniques to adjust grid points based on video input, enhancing accuracy in court analysis.
Implemented three types of projections: Front Wall, Sidewall, and Hit Projection, to map player movements and ball trajectories precisely within the court.
Leveraged the YOLO algorithm for real-time detection and tracking of players, ensuring dynamic analysis of player behavior and positioning.
Employed TrackNet, a CNN architecture tailored for precise object tracking within video sequences, focusing on detecting and tracking sports balls.
Initially developed for badminton shuttlecocks, TrackNet achieved ~78% accuracy in detecting squash balls after being restrained using squash game imagery and data.
By utilizing annotated video data during training, TrackNet effectively learned to recognize and track the ball's trajectory across frames, ensuring accurate detection even in challenging scenarios like occlusions or varying lighting conditions.
TrackNet offers extensive documentation to facilitate seamless retraining or fine-tuning of its base model.
Achieved improved accuracy in grid mapping, facilitating precise analysis of player positions and movements during gameplay.
Successfully mapped player movements and ball trajectories using three distinct projection techniques, providing detailed insights into game dynamics.
Enabled real-time detection and tracking of players on the squash court, allowing for instant analysis and feedback on player performance.
Attained high accuracy in detecting squash balls across frames, even in challenging scenarios such as occlusions or varying lighting conditions.
Our deep understanding of different models and techniques in computer vision, empowered our client to achieve superior performance analysis and efficient gameplay evaluation. By delivering a robust solution we strengthened the position of our client as a pioneer in AI assisted squash analysis.