This website requires JavaScript.

Differentiable Rendering for Pose Estimation in Proximity Operations

Ramchander Rao BhaskaraRoshan Thomas EapenManoranjan Majji
Dec 2022
摘要
Differentiable rendering aims to compute the derivative of the imagerendering function with respect to the rendering parameters. This paperpresents a novel algorithm for 6-DoF pose estimation through gradient-basedoptimization using a differentiable rendering pipeline. We emphasize two keycontributions: (1) instead of solving the conventional 2D to 3D correspondenceproblem and computing reprojection errors, images (rendered using the 3D model)are compared only in the 2D feature space via sparse 2D featurecorrespondences. (2) Instead of an analytical image formation model, we computean approximate local gradient of the rendering process through online learning.The learning data consists of image features extracted from multi-viewpointrenders at small perturbations in the pose neighborhood. The gradients arepropagated through the rendering pipeline for the 6-DoF pose estimation usingnonlinear least squares. This gradient-based optimization regresses directlyupon the pose parameters by aligning the 3D model to reproduce a referenceimage shape. Using representative experiments, we demonstrate the applicationof our approach to pose estimation in proximity operations.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答