University Dissertation Project

Project Duration: 1 Year [Double Module]

For my undergraduate dissertation, I developed a real-time cricket ball-tracking iOS application to make advanced ball-tracking technology accessible to cricket players. Leveraging machine learning with the YoloV2 architecture via CreateML, I trained a model on a meticulously prepared dataset of cricket ball images, enhanced with augmentations for improved accuracy. The application calculates performance metrics based on ball behavior, offering immediate and actionable feedback to players. Developed using agile methodology, the project achieved a precision rate of 86%, though accuracy varied with the ball's distance from the camera. This application provides significant insights for players, helping them enhance their performance in matches.

Image 1


The dashboard in my cricket ball-tracking application presents key performance metrics in a clear and accessible manner, tailored to whether users predominantly bat or bowl. Users can customise the layout by tapping the pencil icon. The bowler's dashboard displays best bowling figures and includes panels for crucial performance metrics, along with a pitch map that visualizes recent bowling sessions. The batsman's dashboard mirrors this structure, showing best batting figures, performance panels, and a wagon wheel visualisation to analyse shot placement and scoring distribution, helping players refine their techniques and strategies.

Image 2


The analytics page in my cricket ball-tracking application, indicated by a cricket ball or bat icon, offers tailored insights for bowlers and batsmen. It features two sections: activity and measurements. The activity section provides "Today," "Week," and "Month" buttons to dynamically update the activity chart, visually representing training frequency and duration. Selecting a bar on the chart updates the measurements section below with relevant session metrics. The measurement section offers customisable metrics tailored to the user's style, with color-coded values matching the pitch map and wagon wheel for coherent and insightful gameplay analysis.

Image 3


The main component of the application, is where real-time ball tracking occurs. This feature captures video footage, analyses each frame, and uses the trained object detection model to track the ball. The visual data is then processed to provide users with analytics displayed within the panel on this page. This page leverages the object detection model I designed in CreateML to deliver accurate and immediate tracking insights.

Image 4


The profile page, allows users to personalise their experience by uploading a profile photo, selecting their team affiliation, and indicating their age. Users can easily fine-tune preferences such as choosing between mph and kph for speed measurement and toggling between feet and meters for distance metrics. A comprehensive guide helps users maximise application features, while a sign-out option provides a straightforward way to conclude their session. The pencil icon in the upper right corner offers a seamless pathway to edit and update personal information.

© 2023 Raaghav Lakhanpal. All Rights Reserved.