G.A. Shadow Classification and Evaluation for Soccer Player Detection.pdf

Shadow Classification and Evaluation for Soccer Player Detection J. Renno, J. Orwell, D.Thirde, G.A. Jones Digital Imaging Research Centre, Kingston University {j.renno, j.orwell, d.thirde, g.jones}@kingston.ac.uk Abstract In a football stadium environment with multiple overhead floodlights, many protruding shadows can be observed originating from each of the targets. To successfully track individual targets, it is essential to achieve an accurate rep- resentation of the foreground. Unfortunately, many of the existing techniques are sensitive to shadows, falsely classifying them as foreground. In this work an unsupervised learning procedure that determines the RGB colour distri- butions of the foreground and shadow classes of feature data is proposed. A novel skelatonisation and spatial filtering process is developed for identifying components in the foreground segmentation that are most-likely to belong to each class of feature. A pixel classification mechanism is obtained at by ap- proximating both classes of feature data by N Gaussian parametric models. To assess our technique’s performance and reliability, a comparison is made with other published works. 1 Introduction Detection of moving objects is essential for automatic monitoring of human activity. A common method for extracting the moving or foreground regions from a video sequence is known as background subtraction [6, 8, 4]. This technique subtracts the incoming video frames from a reference image acquired during a period of inactivity and option- ally updated over time. The resulting pixel or region differences are usually classified as foreground or background by using a statistical or deterministic approach to detect the presence of moving objects in the scene. Shadows cause problems for moving target detection and tracking. The appearance of neighbouring background is changed, to the extent that it can be falsely classified as foreground. Thus, measurements of moving objects are less reliable: this ma