Cricket Ball Trajectory Prediction and Tracking using Hybrid Transfer Learning with ResNet50, Alex Net, ResNet18 and Custom CNN
Keywords:
Custom Convolutional Neural Network, ResNet18, Alex Net and ResNet50.Abstract
This paper introduces a new method for
predicting and tracking cricket ball trajectories, which
is important for analysing player and team
performance. The method uses transfer learning with
well-known convolutional neural network (CNN)
models like ResNet50 and AlexNet, along with a custom
CNN. In the first stage, these models are fine-tuned
using a dataset of labelled ball trajectories to learn the
complex patterns of cricket ball movements. Then, a
combination of object detection and motion estimation
techniques is used to track the ball in video frames. The
experiments, conducted on diverse cricket scenarios,
show that the proposed approach, leveraging pretrained
CNN models, is more accurate than traditional
methods, highlighting its potential in sports analytics
