摘要
注意力视觉是机器人视觉领域的研究热点之一,对机器人智能化发展具有积极意义。为提高机器人交互的自然性,构建了一个基于仿人机器人的注意力视觉选择交互系统。使用图像分割将图像分解为多个特征通道,提取亮度、颜色和方向等多个维度的特征;对不同维度进行多尺度采样,构造高斯金字塔,通过中央——周边差操作提取初级特征图,应用图像融合算法将不同维度的多幅特征图合并得到显著图,通过与低级特征的对比计算位置概率分布偏置,并进行双向加权融合,确定目标位置;设计交互语意实现自然交互。
The attention vision is one of the research hotspots in the field of robot vision, which has positive significance to the intelligent development of robot. In order to improve the naturalness of robot interaction, an interactive visual selection system based on humanoid robot is constructed. First, the image segmentation is used to decompose the image into multiple feature channels. The features of multiple dimensions are extracted, such as brightness, color and direction. Secondly, the multi-scale sampling of different dimensions is carried out. The Gaussian pyramid is constructed and the primary feature graphs are extracted by central-peripheral difference. The image fusion algorithm is used to merge multiple feature graphs of different dimensions to obtain the saliency map. By comparing the low-level features with the location probability distribution bias, the two-direction weighted fusion is used to determine the target position. Finally,the interactive semantics is designed to realize natural interaction.
出处
《计算机与网络》
2017年第11期67-69,72,共4页
Computer & Network
基金
2016年度大学生创新创业训练计划省级基金资助项目(S201612026076/XA201615393)
关键词
注意力视觉选择
图像分割
显著图
仿人机器人
自然交互
attention visual selection
image segmentation
significant figure
humanoid robot
natural interaction
