用改进的深度差分特征识别人体部位  

Improved depth comparison feature for the recognition of human parts

作　　者：张乐锋[1] 郑逸[1] 傅超[1] 

机构地区：[1]浙江工业大学信息工程学院,浙江杭州310023

出　　处：《微型机与应用》2015年第14期54-57,共4页Microcomputer & Its Applications

摘　　要：为了进一步提高人体部位识别正确率,考虑人体部位尺寸不一特性,提出了改进型深度差分特征。改进型深度差分特征根据人体部位尺寸大小确定特征偏移量取值,然后利用随机森林算法训练分类模型,实现了人体部位识别。实验结果表明,采用改进型深度差分特征作为分类模型的训练特征点,实现了人体部位更高、更准确的识别率,比原深度差分特征提高了1.95%。In order to im.prove the correct rate of the recognition of human body parts, considering the different size of human parts, this paper proposes an improved depth comparison feature. The improved feature＇s offset value is decided by the size of human body parts, and uses random forest algorithm to train the model of classification to accomplish the recognition of body parts. The experimental results show that taking the improved depth comparison feature as the classification model＇s training characteristics has a higher and more accurate correct rate of human body parts, increased by 1.95%.

关 键 词：部位识别 随机森林 深度图像 深度差分特征 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]