一种基于SVM的高效球门检测方法

Efficient Goal-mouth Detection Method Based on SVM

为了有效地将足球视频中具有复杂背景的球门检测出来,该文提出一种基于支持向量机(Support vectormachine,SVM)的球门检测算法。首先对于视频图像,利用Top-Hat变换突出白色,得到彩色边缘图像,并对彩色边缘图像灰度化、二值化和形态学连通分析,接着在此基础上提取视频图像中前两根最长并满足一定条件的垂直方向连通的垂线段作为候选球柱,然后计算特征向量,最后利用SVM的强大学习能力进行球门检测。实验证实该方法不仅检测效率很高,而且比已有的球门检测算法有更强的鲁棒性和适应性。

To efficiently detect a goal-mouth with complicated backgrounds in soccer video, this paper proposes a goal-mouth detection algorithm based on support vector machine （SVM）. For each frame, a Top-Hat transform is used to enhance white color and the achieved RGB image is converted to a grayseale intensity image and then converted to a binary image. Based on the morphologic connection analysis, the two longest vertical lines meeting some conditions are achieved and seen as potential goalposts and the feature vector can be computed. With the help of strong study ability from SVM, the goal-mouth is detected. Experiments prove that this algorithm is not only efficient, but also has higher robustness and is more flexible compared with the existing algorithms for goal-mouth detection.