摘要
为了进一步提高行为识别的准确率,将视频中行为的动态特征和静态特征结合起来,应用一种改进的模糊支持向量机(FSVM)方法进行识别,该方法中采用一种新的隶属度确定方法,考虑了样本与类中心的距离以及样本与样本之间的紧密度关系;同时对支持向量机中靠近支持向量的难以识别的样本使用K近邻法识别.在KTH图像数据集上进行实验,将支持向量机与改进的模糊支持向量机两种识别方法进行比较,改进的模糊支持向量机在各类行为识别上取得了较高的识别率.
In order to improve the accuracy in behavior recognition, we dynamic and static characteris- tics of behavior in video were extracted dynamic and static characteristics of behavior in video, and an improved recognition algorithm of fuzzy support vector machine （FSVM） were proposed, while using a new method for the determination of membership degree, considering the distance between samples and the center of the class, also taking the relation of the sample tightness into account. And for the samples near the support vectors that are difficult to identify, the K neighbor method of identification is em- ployed. Experiments on KTH image data sets are performed, and the results using the support vector machine and improved fuzzy support vector machine are compared, the latter method has a higher recog- nition rate.
作者
王向东
张丽红
WANG Xiangdong ZHANG Lihong(College of Physics and Eleetronie Engineering, Shanxi University , Taiyuan 030006, Chin)
出处
《测试技术学报》
2017年第2期125-130,共6页
Journal of Test and Measurement Technology
基金
山西省科技攻关计划(工业)资助项目(2015031003-1)
