轻量化的端到端非合作目标位姿估计方法

Lightweight Network-Based End-to-End Pose Estimation for Noncooperative Targets

针对空间非合作目标六自由度位姿估计问题,基于卷积神经网络设计一种轻量化网络LSPENet,无须手动设计特征便可实现端到端的位姿估计。使用深度可分离卷积、高效通道注意力(ECA)机制构成基本模块,兼顾了网络的复杂度和准确度。设计两个分支分别用于估计位置和姿态,位置估计使用直接回归法,姿态估计引入软分配编码。在URSO数据集的实验结果表明:姿态软分配编码相比直接回归姿态能够显著减小姿态误差;相比其他端到端位姿估计网络,所提网络的参数量减少76.7%,单幅推理时间降低13.3%,同时位置估计精度提高54.6%,姿态估计精度提高57.8%。实现的轻量化端到端位姿估计网络为星载单目视觉位姿估计提供了新思路。

Aiming at the problem of sixdegreeoffreedom pose estimation for noncooperative targets in space,this research involved designing a lightweight network named LSPENet based on convolutional neural networks,which could be used to realize endtoend pose estimation without manually designing features.We used depthseparable convolution and efficient channel attention(ECA)to form the basic module,which balanced the complexity and accuracy of the network.One branch was designed for location estimation using direct regression,and another branch was designed for orientation estimation by introducing softassignment coding.Experimental results on the URSO dataset show that softassignment codingbased orientation estimation exhibits substantially lesser errors than direct regressionbased orientation.Further,compared with the other endtoend pose estimation network,the proposed network reduces parameter count by 76.7%and decreases singleimage inference time by 13.3%,while simultaneously improving location estimation accuracy by 54.6%and orientation estimation accuracy by 57.8%.Overall,LSPENet provides a new idea for monocular visual pose estimation on board.