基于HOG和粒子滤波的足球视频中球的跟踪方法

在足球视频中,由于球员和摄像机的运动使得基于背景的方法无法有效检测和跟踪足球;足球在视频中所占像素个数少、运动复杂,并且在球场中有相似目标如球袜等的干扰,使得目标难以准确跟踪。为改善足球的跟踪效果,文中提出了一种基于梯度向量直方图和粒子滤波的足球跟踪算法。算法以粒子滤波为跟踪框架,用HOG特征与颜色特征融合,结合足球的结构信息与颜色信息,实现了足球视频中足球的稳定跟踪。实验结果表明,文中算法跟踪足球更准确,并可在存在相似目标干扰的条件下有效地跟踪足球。

Football Tracking Algorithm Based on HOG and Particle Filter in Broadcast Soccer Video

In soccer video broadcast,football tracking is very difficult due to the football players' moving,camera zooming,etc.As a consequence,previous background based target detection and tracking method cannot track a football effectively.Football has only a few pixels in an image and moves irregularly and sometimes similar targets such as socks exist,which makes football tracking even harder.This paper presents a football tracking algorithm based on Histogram of Oriented Gradients (HOG) and particle filter to achieve robust football tracking in soccer video broadcast.With the fusing the HOG and color feature,the structure information and color information of a football are well modeled in particle filter tracking framework.The experiment shows that the proposed algorithm is accurate and robust even in the case where similar targets exist.