摘要
有效的动作识别是实现机器人人机交互的关键,在进行人机交互过程中,采用图像识别的方法进行人体动作特征提取和感知,实现动作重构和识别,由于人体动作的随机性,导致大量的干扰动作,需要设计有效的干扰动作排除方法。提出一种基于局部线性嵌入人体动作重构的机器人人机交互中的干扰动作排除方法。建立特征空间背景全局性信息模型,对特征空间中的突变信息进行采集,采用图像处理方法实现对人体动作的特征模型构建,筛选出干扰动作特征,预测人体的动作点,构建全局非显著性二维流形集合矩阵,实现对干扰动作的排除,利用颜色梯度变化对像素点进行局部修正处理,进行干扰动作排除,可以提高对人体动作识别的抗干扰能力。仿真实验结果表明,采用该算法进行人机交互干扰动作排除和识别,能有效提高机器人人机交互的动作识别准确率,抗干扰能力强。
Action recognition effectively is the key to realize the human robot interaction, in the process of interpersonal in-teraction, using image recognition method for human motion feature extraction and perception, action to achieve reconstruc-tion and recognition, due to the randomness of human body movements, resulting in a large number of interference action, need to design effective elimination method of interference action. Put forward the method for eliminating the interference action of human robot interaction locally linear embedding human motion reconstruction based on the. The establishment of characteristic background of global information model, to collect the abrupt information in feature space by using image pro-cessing method, the realization of building a feature model of human action, screening out the interference characteristics of the action, the action point prediction of the human body, the construction of non obvious 2-manifold set matrix realizing global, exclusion of the interference action, in pixels some local correction using the color gradient change, action to ex-clude interference, can improve the anti-interference ability of the human action recognition. Simulation results show that, using this algorithm for the man-machine interactive interference action shooting and recognition, action recognition robot can effectively improve the accuracy of interpersonal interaction, it has strong anti-interference ability.
出处
《科技通报》
北大核心
2015年第10期73-75,共3页
Bulletin of Science and Technology
基金
四川省教育厅科研项目(13ZA0136)
四川民族学院科研项目(12XYZB011)
关键词
机器人
人机交互
动作识别
图像
robot
interpersonal interaction
action recognition
image