Court View Shots Detection Based on Hough Transform and SVM

In tennis match highlights mainly take place in shots containing full court (Court view shots), therefore successful court view shots detection is useful for highlights extraction. This paper proposes a court view shots detection algorithm, in which shot detection that is the precondition of usual shot classification is given up for shot detection not only costs more time, but also its detection performance influences the subsequent shot classification. In this algorithm, a frame is firstly read in from video stream and after a series of image processing, lines are detected based on Hough transform, then eigenvector can be computed and frame containing full court can be detected based on SVM, at last based on time line, continuous frames containing full court and meeting certain conditions are segmented into one court view shot, while other frames are segmented into non-court view shot. Experiments prove this algorithm has satisfying recall and precision.