摘要
由于采摘机器人多视觉目标的识别形式多为独立传感器结构,传感器距离没有被有效约束,导致识别范围受限,快速视觉定位识别单元面对多目标识别过程耗时较长。文章提出一种采摘机器人多视觉目标的快速识别算法,利用视觉传感器收集并标注与采摘目标相关的图像数据,确保数据集涵盖各种目标类型和场景。将识别算法部署在机器人的视觉传感器节点上,在采摘数据集中,检测多个目标及其位置信息,并设置节点间隔距离以满足识别约束要求。利用部署的节点实时采集数据信息,并通过多阶视觉快速定位方法核定采摘机器人定位苹果的位置。对于多个目标重叠在一起而造成识别误差,设计重叠标识修正方法,基于目标特征进行判断和修正,确保每个目标都能够被正确识别。测试结果表明:研究方法能够精准提取果实目标,针对重叠目标也能实现准确识别,且耗时控制在0.4 ms以下,在视觉传感器的辅助与支持下,当前所设计的快速定位识别方法的应用效果更佳,具有实际的应用价值。
Since the recognition form of multi-visual targets of the picking robot is mostly independent sensor structure,the sensor distance is not effectively constrained,resulting in limited recognition range,and the rapid visual positioning recognition unit takes a long time in the process of multi-target recognition.The paper presents a fast multi-visual target recognition algorithm for picking robots.Visual sensors are used to collect and label image data related to the picking target,ensuring that the data set covers a variety of target types and scenarios.The recognition algorithm is deployed on the vision sensor nodes of the robot to detect multiple targets and their location information in the picked data set,and the distance between nodes is set to meet the recognition constraints.The deployed nodes were used to collect data information in real time,and the apple location of the picking robot was verified by multi-level visual rapid positioning method.For the identification error caused by multiple targets overlapping together,the overlapping identification correction method is designed to judge and correct each target based on the characteristics of the target to ensure that each target can be correctly identified.The test results show that the research method can accurately extract fruit targets and achieve accurate recognition of overlapping targets,and the time is controlled below 0.4 ms.With the assistance and support of vision sensors,the current rapid positioning recognition method designed has better application effect and practical application value.
作者
白金柯
BAI Jinke(Henan Technical Institute,Department of Mechanicaland Electrical Engineering,Zhengzhou 450000,China)
出处
《传感器世界》
2023年第10期28-33,共6页
Sensor World
基金
河南应用技术职业学院青年骨干教师资助项目(No.2020-GGJS-J002)
河南应用技术职业学院校级课题项目(No.2022-KJ-48、2023-KJ-62)
河南应用技师职业学院首席技师资助项目(No.2020-SXJS-JD01)
河南省教育厅科学技术重点研究项目(No.22B520016、23B460017)
河南省科技攻关项目(No.222102220099、No.232102210038、No.232102220068)。
