A Unified Framework for Cricket Video Shot Classification using Low Level Features

Abstract

Objectives: Classification of various shots from the cricket video is a fundamental and useful step in cricket video summarization. Methods: We proposed a unified framework for cricket video shots classification. Shots are classified in to Field, Pitch and Boundary, Close-Up, Crowd, Fielders’ gathering and Sky. It requires domain knowledge of cricket sport. Classification of shot into specific category requires extraction of appropriate low level features from the frames. Findings: Multi-perception neural network is then trained and tested with data sets which consist of set of feature vector. Findings: Shot classification accuracy can be increased by adding other features like motion, edge change ratio, texture etc. Basic processing for shot classification is time consuming so parallel approach for the same will be helpful to reduce the execution time. Results: Result analysis shows that system accuracy of shot classification is 75% on an average. Keywords: Classifier, Feature Extraction, Feature Vector, Key Frame, Shot, Training and Testing Data Set