Detection and Recognition of Soccer Ball and Players

Abstract

In this work, an efficient, mere algorithm for detection and recognition of a soccer ball and players has been proposed. It actually involves two stages, detection followed by recognition. In long shot frames, existing algorithms will most likely be unable to identify the ball when it converged with the lines in the field. In this way, we present a strategy that can separate lines from the ball and section the ball capably. First, the ground (background) is disposed of from the scene and the edges of the original frame are distinguished. At that point, Hough line transform is used identify lines. After line identification, the lines are wiped out from the original frame so that just detected players and the ball remain in the scene. Now moving to recognition of player’s. In this work, approach to recognition of players is based on player jersey number. Jersey numbers are stored in database. Jersey number is the base image and the original frame will be the test image. SURF (Speeded-Up Robust Feature) Features in both the base image and test images are found. Then, Features (Feature descriptors) are extracted and match features accordingly. After features getting matched, geometric transform is estimated using Affine Transform. After estimating, the original base image has been recovered, and structural similarity has been checked to avoid false interpretation. If the structural similarity index value is above the defined limit, then the corresponding player statistics will be displayed. Higher the similarity index value, lesser will the probability for false interpretation