Research On The Billboard Detection In Sports Video

In sports events,billboard is an important source of income.Billboard detection is a significant part of virtual advertisement system.In consideration of the variability of illumination,materials,occlusion and distribution,it's difficult to detect billboards of different scenes.In this paper,billboard detection and location,which aims at improving the general applicability of the algorithm,is studied.The main contents of the paper are as follows:1.This paper presents a detection algorithm based on court features and saliency map.Billboard region in various special scenes is located by using dominant color extraction,b-spline curve fitting,and the fusion of saliency map and distance statistics.Compared with several common saliency segmentation algorithms,the algorithm improves the general applicability.2.This paper presents graph cuts algorithm based on background model.The image which is removed from the playfield region is used to establish spatial background model in HV channel.The background model is iterated by temporal data.The frame difference method decides whether to re-initialize the background model or not.The thesis solves the minimum energy of Gibbs function by the maximum flow method to detect the billboard region.B-spline curve fitting and height statistics are used to modify the billboard boundary.Compared with other common image segmentation algorithms,this algorithm can get higher accuracy in the special scenes,which makes up for the deficiency of some algorithms and improves the general applicability of the algorithm.3.A tracking algorithm based on optical flow is proposed for the billboard region tracking problem.The feature points,which are extracted by SURF algorithm,are tracked by optical flow and used to calculate the homography matrix,then the RANSAC algorithm is used to continuously modify the model.The thesis detects billboard on next frame by tracking the border of billboard region.Compared with the typical point matching algorithms,the algorithm has a shorter running time.In order to lay a foundation for the following work,the thesis compares three billboard detection algorithms and analyzes the advantages and disadvantages of the three algorithms.