摘要
本文提出了一种局部二值模式(Local Binary Pattern,LBP)特征和运动特征相融合的运动分类算法。首先,利用背景差分法检出视频中的运动人体序列,将运动人体序列经过LBP算子处理得到LBP直方图特征。然后,将LBP直方图特征和运动人体质心的速度特征相融合作为运动人体行为分析的识别特征,应用BP神经网络进行行为分类识别。在Weizmann和KTH行为数据库上进行了算法实验研究,人体行为识别的平均准确率达到了90.78%。实验结果表明:该方法在识别率方面明显优于常规方法进行识别的结果。
This paper presented an classification Algorithm based on multi-features fusion using Local Binary Pattern（ LBP） and motion features. Firstly,background subtraction algorithm is used to extract the human motion sequence in the video,and use the LBP operators to calculate the samples’ LBP histogram feature. Then,combine LBP histogram feature with the speed feature of the moving human body’s centroid,and take it as recognition feature of the human movement analysis. At last,using BP neural network for action classification and recognition. The experimental study of the proposed algorithm is carry out on the Weizmann behavior database and the KTH behavior datebase,and the average accuracy rate of the human action recognition is 90. 78%. The experiment result shows that the above-mentioned method is superior to conventional method for recognition.
出处
《激光杂志》
CAS
北大核心
2015年第2期56-59,共4页
Laser Journal
关键词
行为识别
特征提取
局部二值模式
BP神经网络
Action recognition
Feature extraction
Local Binary Pattern
BP neural network
