摘要
提出了一种新的运动目标分类方法,该方法利用sigmoid函数将标准SVM的输出结果直接转换为目标所属类别的概率,避免了分类器的组合问题;同时该方法还利用后置加权均值滤波器对SVM的初始输出结果进行滤波平滑处理,进一步提高了分类的正确率。实验结果表明,该方法能有效地提高运动目标分类的精度。
This paper presented a new method to classify moving targets, in which the outputs of standard SVMs could be mapped directly into target category' s posterior probabilities by the sigmoid function. Furthermore, also put forward a post-filtering framework to improve classification accuracy, using a weighted average filter to smooth the initial outputs of SVM classitiers. Experimental results demonstrate that the framework of SVM probability outputs combined with a post-filter is more effective for moving target classification from video in terms of classification accuracy.
出处
《计算机应用研究》
CSCD
北大核心
2010年第2期778-780,共3页
Application Research of Computers
基金
重庆市自然科学基金资助项目(CSTC-2008BB2252)
国家大学生创新性实验计划资助项目(081063510)
关键词
支持向量机
后验概率
均值滤波
运动目标分类
SVM (support vector machine)
posterior probability
average-filtering
moving target classification