摘要
行为识别技术是一种可用于智能视频监控的生物识别技术。本文提出一种基于信息论中相对熵概念的行为识别算法。训练过程使用Parzen方法估计概率分布函数,测试过程则将图像序列投影到低维特征空间,使用相对熵作为判断两种不同行为的相似度函数,在Weizmann数据库上测试的结果验证了算法的有效性。
Action recognition technology is a bio-identification technology which can be used for intelligent video surveillance system.In this paper,the problem of representing human actions using visual cues for the prose of learning and recognition was addressed.Parzen window was used to estimate the probability distribution function in training stage.Supervised pattern classification with relative entropy as distance measure function is finally performed in the lower-dimensional for recognition under projection space.Extensive experimental results on Weizmann databases reveal that proposed algorithm has an encouraging recognition performation.
出处
《河南科技大学学报(自然科学版)》
CAS
北大核心
2009年第6期53-55,共3页
Journal of Henan University of Science And Technology:Natural Science
基金
国家自然科学基金项目(70781043)
关键词
相对熵
行为识别
模式分类
流行嵌入
维数约减
Relative entropy
Action recognition
Pattern classification
Manifold embedding
Dimensionality reduction