摘要
交互物体的检测识别是实现人机交互的一项关键技术,针对人机交互过程中交互物体检测范围受限的问题,本文利用深度归一化提高深度图像质量,提出了一种基于图像分割的任意交互物体检测方法。该方法针对操作人员侧向和正向姿态,分别采用基于显著性检测的图像处理和人体姿态引导的区域生长算法分割目标区域,锚定目标物体边框实现物体检测。最后,进行了交互物体检测实验及不同深度区间位置测距和跟随实验。实验结果表明,所提出的物体检测方法能够实现任意交互物体检测,在交互物体检测方面具有广泛适用性;较小深度区间的归一化能够使物体位置误差变小,提高了物体检测距离精度及机器人跟随效果。
Detection and recognition of interactive objects is a key technology to realize human-computer interaction,and in order to solve the problem of limited types of interactive objects in the process of human-computer interaction,an arbitrary interactive object detection method was proposed based on image segmentation.Firstly,for the depth image,after filtering out the data outside the range of the original depth data,the min-max scale normalization method was used to improve the quality of the depth image.Secondly,the target area was segmented by using the image processing method based on saliency detection and the human pose-guided region growth algorithm for the operator's side-to-side camera and front-facing camera posture,respectively.Then,the pixel set of the target object obtained by the above segmentation was input into the image processing functions,and the minimum external rectangle of the area point set was obtained,and the rotating bounding box of the target object was anchored.Then,for the depth image,after filtering out the data outside the range of the original depth data,the min-max scale normalization method was used to improve the quality of the depth image.Finally,the detection experiments of arbitrary interactive objects and the ranging and following experiments of different depth intervals were carried out.Experimental results showed that the proposed object detection method had a lower detection cost and a higher degree of freedom in the detection category of interactive objects,which can realize the detection of arbitrary interactive objects,and had wide applicability in the detection of interactive objects.The normalization of the small depth interval can effectively improve the depth image quality,make the object position error smaller,and improve the accuracy of the object detection distance and the following effect of the robot in the human-computer interaction experiment.
作者
黄玲涛
孔紫静
杨帆
张红彦
HUANG Lingtao;KONG Zijing;YANG Fan;ZHANG Hongyan(School of Mechanical and Aerospace Engineering,Jilin University,Changchun 130022,China)
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2024年第8期428-436,共9页
Transactions of the Chinese Society for Agricultural Machinery
基金
吉林省重点研发计划项目(20200401130GX)
国家自然科学基金项目(51575219)。
关键词
目标物体检测
深度归一化
图像分割
人机交互
target object detection
depth normalization
image segmentation
human-computer interaction
作者简介
黄玲涛(1979-),男,副教授,主要从事机器人控制和主从控制及力反馈技术研究,E-mail:hlt@jlu.edu.cn;通信作者:张红彦(1978-),女,副教授,主要从事机器人技术和机器视觉处理研究,E-mail:Zhanghy@jlu.edu.cn。
