Aiming at the problem that the relative position measurement range of intelligent ammunition to mili-tary 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 inter-val 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 non-linearly map the position parameters to the(0,1)interval to improve measurement accuracy and stability.Ex-perimental 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.
维普
万方数据