论文部分内容阅读
提出了一种结合词袋法的3维尺度不变特征转换(3D-SFIT)算子,并应用于人的动作识别.将运动的人从图像背景分割出来并计算特征算子,用词袋法表征视频,最后采用支持向量机(SVM)对动作进行分类.采用Weizmann动作数据库对本方法进行测试,实验结果表明:3D-SFIT算子能很好地描述视频序列的本质,比传统的描述算子更为高效;同时能更好地适应光照变化和施动者的穿着和动作差异等环境因素的影响,取得更好的识别效果.
A 3-dimensional scale-invariant feature transformation (3D-SFIT) operator combined with the word bag method is proposed and applied to human motion recognition. The moving person is segmented from the image background and the feature operator is calculated. Method is used to characterize the video.At last SVM is used to classify the motion.Weizmann motion database is used to test the method.The experimental results show that 3D-SFIT can describe the essence of video sequence better than traditional description arithmetic Child more efficient; at the same time better able to adapt to light changes and the actors who wear and movement differences and other environmental factors, and achieve better recognition results.