基于深度变换的三维位置测量方法  

A 3D Position Measurement Method Based on Depth Transform

作　　者：郝现伟 翟晋 吴姝源 王泽鹏 HAO Xianwei;ZHAI Jin;WU Shuyuan;WANG Zepeng(Beijing Institute of Aerospace Systems Engineering,Beijing 100076,China;Beijing Institute of Technology,Beijing 100081,China)

机构地区：[1]北京宇航系统工程研究所,北京100076 [2]北京理工大学,北京100081

出　　处：《探测与控制学报》2024年第6期58-64,共7页Journal of Detection & Control

摘　　要：针对智能弹药对军用车辆的弹目相对位置测量范围大,通过神经网络直接回归该参数时精度差、稳定性低等问题,提出一种基于深度变换的目标三维位置测量方法。该方法将目标的三维位置测量问题转化为目标几何中心点在图像中的位置估计和深度估计问题;以YOLOv5网络为算法框架,采用区间非线性映射和有序分类代替回归2种策略,解决直接回归难度较大的问题,并保证在不同距离下测量相对误差稳定;利用Sigmoid函数将位置参数非线性映射到(0,1)区间来提高测量的精度和稳定性。实验结果表明,测量距离在50~400 m范围内时,位置的相对测量误差小于5%,满足测量需求。Aiming at the problem that the relative position measurement range of intelligent ammunition to military vehicles is large,and the accuracy and stability are poor when the parameter is directly regressed by neural network,a 3D position measurement method based on depth transform was proposed.The method transformed the 3D position measurement problem into the position estimation and depth estimation of the target’s geometric center point in the image,and took YOLOv5 network as algorithm framework,and then two strategies of interval nonlinear mapping and ordered classification were used to solve the problem of difficult direct regression and ensure the stability of measurement relative error at different distances,which adopted Sigmoid function to nonlinearly map the position parameters to the(0,1)interval to improve measurement accuracy and stability.Experimental results showed that when measurement distance within the range of 50 m to 400 m,the relative measurement error of position was less than 5%,which met measurement requirements.

关 键 词：智能弹药 位置测量 深度变换 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]