Content-based cricket video shot classification using bag-of-visual-features

Abstract

In this paper, we propose a novel method for content-based cricket video shot classification using bag-of-visual-features. Bag-of-visual-features methodology has attained huge popularity among the computer vision community. The standard steps followed in bag-of-visual-features method are unordered collection of set of local features from training images and clustering them to form a visual vocabulary which is useful for training and classification. Bag-of-visual-features technique is the best choice for content-based indexing and classification due to its simplicity and accuracy. Cricket is having huge popularity throughout the world. Cricket video consists of different types of shots (a shot can be considered as continuous sequence of frames recorded between the start and stop of a single camera. For example, different video shots present in a cricket video are close-up view shots, distance view shots, and pitch view shots. Our proposed method is based on bag-of-visual-features for classification of cricket video shots by using top 100 SIFT (Scale-Invariant Feature Transform) features selected from each frame of the training and testing video shots. We have used the dataset of cricket, made available online (Video dataset, http://cse.iitk.ac.in/~vision/dipen/) by Mr. Dipen Raghuwani, to evaluate our proposed framework. By using our proposed framework, we are able to achieve reasonable results compared to already existing methods.