Ball recognition in images for detecting goal in football

摘要

In this paper we present a technique for detecting goals during a football match by using images acquired by a single camera placed externally to the field. The method does not require the modification either of the ball or of the goalmouth. Due to the attitude of the camera with respect to football ground, the system can be thought of as an electronic linesman which helps the referee in establishing the occurrence of a goal during a football match. The occurrence of the event is established detecting the ball and comparing its position with respect to the location of the goalpost in image. The ball detection technique relies on a supervised learning scheme called Support Vector Machines for classification. The examples used for training are appropriately filtered version of views of the object to be detected, previously stored in form of image patterns. We have extensively tested the technique on real images in which the ball is both fully visible and partially occluded. The performance of the proposed detection scheme are measured in terms of detection rate, false positive rate and precision in the ball localization in image.