采摘机器人的目标定位和测量技术

Target localization and measurement technology for picking robots

在采摘机器人的工作过程中,为了提高采摘机器人的采摘成功率,需要获取水果的位置信息,以确定果实与采摘机器人的相对位置关系.由于采摘作业环境复杂,为提高采摘系统的工作效率,提出一种基于Opencv采用Yolov5算法和双目相机对水果进行目标识别和空间定位的方法.针对小目标识别在Yolov5算法中识别精度的不足,在Yolov5算法网络结构中叠加包含更多低层级信息的浅层特征图,实现小目标检测层进行算法优化,实验结果表明,优化后的识别网络对水果检测的平均精度为92.4%.基于深度学习的优化识别网络在识别小目标方面具有更好的性能,可以有效提高果农采摘系统的工作效率.

During the works of the picked robot,in order to improve the picked robot′s picked success rate,the position information of the fruit needs to be obtained first to determine the relative position relationship between the fruit and the picked robot.Due to the complex environment of the picked operation,a method of target recognition and spatial localization of fruits based on Opencv uses Yolov5 algorithm and binocular camera was proposed to improve the efficiency of the picked system.For the shortage of recognition accuracy of small target recognition in the Yolov5 algorithm,a shallow feature map containing more low-level information was superimposed on the network structure of the Yolov5 algorithm to achieve an algorithm optimization of the small target detection layer,and the experimental results showed that the average accuracy of the optimized recognition network for fruit detection was 92.4%.The optimized recognition network based on deep learning has better performance in identification of small targets,which can effectively improve the efficiency of fruit picked system.