摘要
针对基于局部时空特征与特征词袋模型相结合的算法中计算效率低、识别率不高的问题,提出一种基于局部块模型与特征数据预处理结合的行为识别算法。该算法基于特征词袋模型,采用局部块模型提取特征,将多变量的复杂问题简化为低维空间的简单问题,对其数据处理过程进行改进,同时优化了局部块模型中的帧采样过程。与原算法相比,计算效率与识别率都有较大提升,较其他同类算法也具有一定优势。2个通用视频库上的实验证明了算法的有效性。
In order to solve action recognition system's efficiency and accuracy in complex environment,an improved method based on local part model and feature pre-processing was proposed. The algorithm was based on the bag of features,and local part model was used to extract features. The features were pre-processed through principal component analysis methods,meanwhile the local part model of frames in the sampling process was optimized. Experiments were carried on two benchmark datasets,named UCF Sports and HMDB51,respectively. The results showed that this algorithm had higher efficiency and accuracy than the original in complex environment. Compared with other methods,the proposed algorithm showed more accurate and efficient.
出处
《四川大学学报(工程科学版)》
EI
CAS
CSCD
北大核心
2016年第S2期121-126,共6页
Journal of Sichuan University (Engineering Science Edition)
基金
成都市科技惠民项目资助(2015-HM01-00293-SF)
特殊环境机器人技术四川省重点实验室项目资助(14zxtk03)
国家自然科学基金委员会和中国工程物理研究院联合基金资助项目(11176018)
关键词
行为识别
局部块模型
特征预处理
特征词袋模型
action recognition
local part model
feature pre-processing
bag of features