Classification of Shot Selection by Batsman in Cricket Matches Using Deep Neural Network
MetadataShow full item record
Machine learning (ML) is such a field that focuses on learning based method. It basically leverage data to improve the performance on particular tasks. It creates a model based on training data and makes prediction according to the pattern what it has learnt. Machine learning can be used to classify a certain category of image as it has a successful contribution in image processing. That’s why we have used machine learning approach to implement our proposal. Our proposal is basically classification based. As we know cricket is a very popular game in our country. Technological advancement has brought a tremendous change in field of cricket. Such as, projected score prediction, wicket prediction, winning probability, run rate as well as shot detection also it has benefitted the decision making system a lot. Our primary objective is to use Machine learning in the field of Cricket, where we aim to classify the tentative shot selection of batsman. Our primary goal is to automate the broadcast system where cameras can move automatically by identifying the shots and the direction of the shots. As sometimes the shots are delivered so fast, crucial moments can be missed due to lack of fast telecast system. For implementing our proposed model, we have generated our own dataset named “CrickShots” by taking real time photos from various cricket matches. We collected 1800 images of batsman while delivering the shots or to be more specific we have tried to take pictures of the connection moment of the bat and ball. To have an accurate result of classification we have used ‘VGG-16’ model and ‘Inception’. Where we got a better result by using VGG-16. We have used 85% of the total images to train the model first and 15% later on to test the model. The images had to go through several pre-processing methods such as background removal and scaling to be prepared for training the model. At last we got desired accuracy of 95% from VGG-16 and 85% from Inception.