No-ball advisory system for bowlers with foot-crease distance estimation

We’ve developed a new system to help cricket bowlers during net practice by quickly spotting whether a delivery is legal or a no-ball. It’s a deep learning-based approach that uses YOLOv8 to detect the bowler’s foot and the crease line, Inception V3 to classify the delivery, and Euclidean distance to measure how far the foot is from the crease. We trained it on a dataset of 2,225 images and tested it on 265 more, using Roboflow for labeling and adding synthetic images to make the model more adaptable. The results are promising: we achieved a precision of 0.92 for legal deliveries and 0.89 for no-balls, with solid recall scores too. Through a Streamlit interface, bowlers get visual feedback right away, which helps them adjust their technique on the spot. This tackles the drawbacks of traditional coaching–like inconsistent feedback–and opens doors for future improvements, such as analyzing images from multiple angles, adding audio-visual alerts, and deploying the system on edge devices for broader use. Our work shows how AI can make a real difference in sports training, offering a practical tool to boost cricket performance.