一种新的基于多信息融合的智能控制方法  

New Method of Intelligent Control Based on Multi-Information Fusion

作　　者：孙瑜欣 SUN Yuxin(Shanghai Institute of Process Automation and Instrumentation,Shanghai 200233,China)

机构地区：[1]上海工业自动化仪表研究院有限公司,上海200233

出　　处：《自动化应用》2024年第7期39-41,45,共4页Automation Application

摘　　要：提出一种新的融合人脸与人体姿态识别的智能控制方法。该算法首先通过人脸识别进行身份鉴别,然后基于深度图像采集人体骨骼信息进行姿态识别,最后根据识别的姿态信息进行指令重构。其中,人脸识别部分采用较新的LTDP2算法(局部自适应三值微分模式)。经实验验证,LTDP2算法能够高效地提取人脸特征,在复杂光线场景下仍可以保持较高的识别准确率。姿态识别部分采用三目摄像头采集人体深度信息,通过对骨骼点相对位置的分析,获得姿态指令,然后对单一指令进行组件化重构,最终拼接成一套完整的控制指令。该算法能够有效解决在封闭无人的场景下对智能系统进行控制的问题。This paper advances a new method for intelligent control which based on face and human posture recognition.The algorithm first performs identity authentication through face recognition,then collects human bone information based on depth images for gesture recognition,and finally performs instruction reconstruction based on the recognized gesture information.Among them,the face recognition part adopts the newer LTDP2 algorithm(second-order local ternary derivative pattern).Experiments have verified that the LTDP2 algorithm can efficiently extract facial features,and can still maintain a high recognition accuracy rate in complex light scenes.The posture recognition part adopts the trinocular camera to collect the depth information of the human body,obtains the posture instruction through the analysis of the relative position of the bone points,then form a complete set of control instructions by componentized refactoring the single instruction.This algorithm can effectively solve the problem of controlling intelligent systems in closed unmanned scenarios.

关 键 词：人脸鉴别 局部自适应三值微分模式 深度图像 姿态识别 指令组件化重构 智能控制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]