基于深度卷积神经网络的激光跟踪仪高精度位姿数据提取  

High precision pose data extraction with laser tracker based on deep convolutional neural network

作　　者：包玉 BAO Yu(Rongzhi College of Chongqing Technology And Business University,Chongqing 401320,China)

机构地区：[1]重庆工商大学融智学院,重庆401320

出　　处：《激光杂志》2021年第9期190-194,共5页Laser Journal

基　　金：国家自然科学基金青年科学基金(No.21805036);重庆市教委科技处项目(No.KJQN201802102)。

摘　　要：传统的激光跟踪仪高精度位姿数据提取方法受到激光跟踪仪自身误差的影响,提取精度严重不足。为此,提出基于深度卷积神经网络的激光跟踪仪高精度位姿数据提取方法。设计数据采集电路,采集目标二维坐标、光线长度、角度数据等原始数据,为不同位姿数据传输要求,设计LVDS通信接口,确保数据的稳定传输,将获得的数据输入到经过训练后的神经网络中,其卷积核通过损失函数计算得到,卷积神经网络最终输出的结果即为高精度位姿数据。实验结果表明:所提方法变量之间线性度好、位姿数据误差极小,数量级达到0.000 1,该方法的提取精度能够满足用户实际应用需求。Conventional methods for extracting high-precision attitude data from laser trackers are affected by the error of the laser tracker itself,and the extraction accuracy is seriously insufficient. Therefore,a deep convolutional neural network based method for extracting high-precision position attitude data from laser trackers is proposed. Design the data acquisition circuit to collect the original data such as target two-dimensional coordinates,ray length,angle data,etc.,and design the LVDS communication interface to ensure the stable transmission of data for different positional data transmission requirements. The obtained data are input to a trained neural network,whose convolutional kernels are computed by a loss function,and the final output of the convolutional neural network is the high-precision positional data. The experimental results show that the proposed method has good linearity between variables,minimal pose data error,and the order of magnitude is 0. 000 1 degrees. The extraction accuracy of this method can meet the actual application requirements of users.

关 键 词：深度卷积神经网络 激光跟踪仪 高精度 位姿数据 数据提取 

分 类 号：TN249[电子电信—物理电子学]