Goal-Bound Soccer Ball Positioning from Uncalibrated Video Sequences

摘要

In these paper we examine the problem of object positioning in the 3D Euclidean space from uncalibrated images. It is well known that this can be accomplished if partial informations about some points or planes in the observed scene arc available. We consider the specific context of plotting the 3D position of a goal-bound soccer ball. In this context we know in the 3D Euclidean space the positions of the corners of the six-yard area and the goal-posts. In literature, a first attempt to solve this problem from two stereo image sequences was based on the vertical projection of the ball on the ground plane and 3D positioning with respect the goal line. In this paper we consider a new approach based on the intersection of view lines estimated by considering the two known planes of the play-ground and the goal-door. We compare the performances of these two approaches on different simulated situations. Moreover we also apply our method without considering the known information about the scene, but introducing artificial landmarks, that are more likely to be recovered in the images, on ideal virtual planes, in order to determine which is the minimum error can be performed in this application context. The results obtained on the simulated data have been verified on some real images acquired with two TV cameras appropriately positioned in the real environment.