New approach for automatic view detection system in tennis video

Abstract

Tennis is the highest structured sport and the playing time is the most important part. It is frequently characterised by a global court view. So, to extract the playing time, we must detect court view frames that represents a significant landmarks in tennis broadcasting. In this context, this paper presents a new system for automatic view detection in tennis video based on pattern recognition approach. First, we detected and extracted the pattern presenting the court view. Then, each current frame of the video is compared to the pattern to yield a colour similarity descriptor. We used measures similarity, based on Hue-histogram, between the current frame and the court view pattern. In addition, we presented an evaluation of our proposed system and a comparison between the different metrics of similarity used (intersection, correlation, chi-square and Bhattacharyya). Experimental results proved the efficiency of our proposed approach in terms of precision and speed.