Automated Highlight Generation from Cricket Broadcast Video

This paper presents a novel method for the automation of highlight extraction using broadcast cricket video. The top-down hierarchical approach yielded an average frame processing speed of 0.04 seconds. The Motion History Image (MHI) method was used to detect the camera zoom-in motion which is a semantic feature of the bowler run-up sequence. A multi-spatial approach to feature extraction maximized highlight detection accuracy and was useful for smart parsing through video sequences. Experimental results on various broadcast video samples showed a robust performance across different formats of the game with an average recall rate of 99% and precision rate of 94.2 %. The proposed framework does not require any supervised training, temporal reordering of frames or manual intervention during the highlight extraction process.