Detection of Cricketing Activities Using Deep Learning

This paper approaches an image classification which is based on cricketing activities with the help of Convolutional Neural Network. Fundamentally the aim is to detect the diverse incidents which come to pass throughout a cricket match. It includes different cricket shots played by the cricketers, several symbols which are betokened by the umpires. This scheme has compelled us to develop a distinct absolute dataset Cricket Image Classification (CIC), consisting of various cricketing incidents. Due to the scarcity of ready images, this paper utilizes transfer learning. This paper approaches a custom ResNet50 model together with the utilization of One Cycle Policy to make a robust and time efficient model. For comparison and to provide a better understanding, four other customized models namely ResNet34, ResNet101, VGG16 with Batch Normalization and AlexNet are introduced and studied in this paper. The model is so efficacious that it can even detect incidents in the quaint ambiance of street cricket, although the model is trained with the images of fields and stadiums in the dataset. The final model achieves 99.47% accurate test results.