KUCricShot: A Dataset for Developing Detection Model of Cricket Shots

Extensive research is necessary to automate various aspects of cricket. Currently, live cricket commentary and the analysis of batters’ posture both require manual intervention. Automating the event detection during game time would be the initial step; we suggest automating the detection of cricket shots. In this stage, due to the shortage of a noteworthy effective public dataset, we’ve focused our attention towards making a dataset. A collection of 1278 videos with four types of cricket shots have been accumulated. Furthermore, upon reviewing prior research, we’ve identified deep-learning tools as the means to a) develop a shot detection model and b) prove the efficacy of our dataset. A comparison-based study is presented in this paper with Convolutional Long Short-Term Memory (ConvLSTM) and the Long-term Recurrent Convolutional Network (LRCN) for the above-mentioned causes. The LRCN model gave us a better result with a validation accuracy of 73%. We anticipate that this dataset will be validated as valuable for the upcoming research works.