基于深度图像学习的人体部位识别  

Human Body Part Recognition Based on Depth Image Learning

作　　者：林鹏[1] 张超[1] 李竹良[1] 赵宇明[1] 

机构地区：[1]上海交通大学自动化系系统控制与信息处理教育部重点实验室,上海200240

出　　处：《计算机工程》2012年第16期185-188,共4页Computer Engineering

基　　金：国家自然科学基金资助项目"多视角下的多类型目标识别与行为分析"(61175009)

摘　　要：针对人体部位识别问题,提出一种基于深度图像学习的人体部位识别系统。构建深度图样本库,包括训练集和测试集,提取训练样本中的局域梯度特征,利用随机森林学习得到分类器,并对图像进行单点分类,计算人体各关节点。实验结果表明,该系统能快速准确地识别人体的不同部位。Aiming at human body recognition problem, this paper proposes a human body part recognition system, which is based on depth image learning. It constructs depth image sample library, including training set and testing set, extracts local gradient feature from training samples, uses random forest to learn the classifier for separating each single points going through the image and computes each joint point of human body. Experimental result shows that the system can recognize different human body parts fast and accurately.

关 键 词：人体部位识别 深度图像 随机森林 监督学习 局域梯度特征 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]