Detection of Bowler Errors using Deep Learning

Cricket is a bat and ball sport played between two teams of eleven players each. There are two innings in a game for each team, and when one team takes the fielding, two players from the other team will get the chance to bat. There are 4 umpires for giving perfect decisions. Two umpires are on the ground and the third umpire looks all things via computerized cameras. T20, test, ODI’s are the some of the different types in cricket matches. It is the most popular and widely played game in the world and hence the decisions must be accurate. In every cricket match there are many different scenarios to classify legal and illegal deliveries. No-Balls, wide ball, bounce etc. are the most common umpire decisions in every cricket match. The classification of these decisions plays an important role in a fair game. Earlier research on automated third umpired decision was done using text dataset with accuracy of 80%. We propose a model on automation of third umpire’s decision using image dataset. In the existing system third umpire has to give the decisions using the technologies like Decision Review System, ultra edge, hotspot. This will take more time to give decisions and needs an expert also to display the decisions. In such cases the whole process can be automated and can increase the transparency in decisions within less time. We will be using images that contains different required scenarios. Since there is no standard dataset available, we created our own dataset that includes 20 bounce, 24 no ball and 10 wide ball images. We will be using CNN and VGG16 algorithms and will make comparison between the accuracies obtained from them. This model can provide spontaneous decisions without the human intervention. The proposed model has an accuracy of 99.7% for CNN and 57.6% for VGG16.