Title :
Automated highlight generation from cricket broadcasts using ORB
Author :
Daniel Ringis;Akash Pooransingh
Author_Institution :
Department of Electrical and Computer Engineering, The University of the West Indies, St. Augustine, Trinidad and Tobago
Abstract :
This paper investigates the use of the Oriented Fast, Rotated Brief (ORB) method to automatically detect the most significant broadcast view associated with cricket broadcasts: the Bowler Run-up Sequence (BRS) for cricket highlight generation. This method is computationally less expensive than other methods proposed for BRS detection. It is shown here that only a single frame is required for training to produce acceptable results as compared to other BRS detection methods. The ORB method produced a BRS detection rate of 98.26% and false detection rate of 5.70% and average matching time of 0.059 seconds. The fast training and matching makes this ideal for real time generation of highlight sand the detection is robust to sequences with high camera motion.
Keywords :
"Training","Feature extraction","Cameras","Robustness","Motion estimation","Real-time systems","Support vector machines"
Conference_Titel :
Communications, Computers and Signal Processing (PACRIM), 2015 IEEE Pacific Rim Conference on
Electronic_ISBN :
2154-5952
DOI :
10.1109/PACRIM.2015.7334809