Beyond the Naked Eye — Automated Delivery Analysis in Cricket

<header>
    <h1>Beyond the Naked Eye — Automated Delivery Analysis in Cricket</h1>
    <p><strong>Project Link:</strong> <a href="https://ieeexplore.ieee.org/document/11333604" target="_blank" rel="noopener noreferrer">https://ieeexplore.ieee.org/document/11333604</a></p>
</header>

<main>
    <section id="project-details">
        <h2>Project Details</h2>
        <ul>
            <li><strong>Overview:</strong> Engineered a low-cost, single-camera 3D ball trajectory tracking and delivery length classification pipeline for cricket coaching applications, published as first-author research at the IEEE FIT 2025 conference (15.7% acceptance rate)[cite: 1, 3, 5, 7].</li>
            <li><strong>3D Trajectory Tracking:</strong> Designed a custom computer vision pipeline utilizing YOLOv5 for real-time ball detection and Kalman Filters (KF) for motion estimation and continuous occlusion-robust tracking[cite: 1, 3, 5, 7].</li>
            <li><strong>Homography Translation:</strong> Implemented camera-to-pitch perspective mapping (Homography) to convert 2D video pixel coordinates into 3D metric spatial data on a standard 22-yard pitch[cite: 1, 3, 5, 7].</li>
            <li><strong>Delivery Length Classification:</strong> Automated metric analysis of ball bounce location to categorize deliveries (e.g., Short, Good, Full, Yorker) with 93% classification accuracy, significantly outperforming legacy single-camera approaches[cite: 1, 3, 5, 7].</li>
            <li><strong>Benchmarking &amp; Accuracy:</strong> Validated system performance with a Mean Absolute Error (MAE) of ~2.5 cm for bounce point mapping, creating a highly viable alternative to expensive dedicated multi-camera setups[cite: 1, 3, 5, 7].</li>
            <li><strong>Technical Stack:</strong> Built using Python, PyTorch (YOLOv5), OpenCV, Kalman Filters, Image Processing, and NumPy[cite: 1, 3, 5, 7].</li>
        </ul>
    </section>

    <section id="short-description">
        <h2>Short Description</h2>
        <p>First-author IEEE Computer Vision paper (15.7% acceptance) automating 3D ball tracking &amp; trajectory classification in cricket with 93% accuracy using YOLOv5[cite: 1, 3, 5, 7].</p>
    </section>

    <section id="skills">
        <h2>Skills</h2>
        <p>Computer Vision, Machine Learning, Artificial Intelligence, Python, Research &amp; Development, Image Processing, Object Detection[cite: 1, 3, 5, 7]</p>
    </section>
</main>