面向灵巧操作的视觉目标识别  

Visual object recognition for smart manipulation

作　　者：雷丽充 刘华平[2] 孙富春[2] 高蒙[3] 张军惺[3] 

机构地区：[1]中铁十七局集团有限公司勘察设计院,山西太原030032 [2]清华大学计算机科学与技术系智能技术与系统国家重点实验室,北京100084 [3]石家庄铁道大学,河北石家庄050043

出　　处：《智能系统学报》2015年第1期37-42,共6页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金重大国际合作研究资助项目(61210013);国家自然科学基金资助项目(60621062,90716021)

摘　　要：为了能够实现灵巧手对目标物体进行精准操作,研究了一种利用Kinect检测出目标物体,在帧差法的基础上对获取的深度进行背景相减,获取出目标物体的运动点,在此基础上利用获取的目标物体的特征采用T-S模糊逻辑判断出目标物体的方法,以BH8-280对目标物体进行抓取实验为例,在实验中,Kinect在帧差法的基础上检测出目标物体的位姿,大小,形状,以此为基础建立起T-S模糊逻辑系统,判断出目标物体的属性和类别,通过实验结果进一步说明了利用本文研究的方法显著地提高了判断物体的准确率和效率,为灵巧手的精细控制抓取奠定了基础。In order to achieve fine operation of dexterous hand,a new method is proposed. The method consists of detecting a targeted object by Kinect,conducting background subtraction for the obtained depth on the basis of frame difference method to acquire the motion point of the targeted object and judging the targeted object by the TS fuzzy logic along with the features of the obtained targeted object. This paper uses BH8-280 for an object grasping experiment to verify the accuracy and rapidity of the method. In this experiment the pose,size and shape of the objects can be determined. The property categories of objects are judged by the T-S fuzzy logic system. It is seen that the dexterous hand can grasp the object accurately. The research of this method builds a foundation for fine operation of a dexterous hand.

关 键 词：KINECT 视觉 背景相减法 帧差法 T-S模糊逻辑 目标识别 精细控制 灵巧操作 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]