多尺度特征融合室内三维场景快速重建
Fast reconstruction of indoor 3D scene based on multi-scale feature fusion
杨周花1
作者信息
摘要
传统三维场景重建方法多是采用激光扫描技术获取结构数据,由于采集设备对环境空间的要求,室内图像采集过程极易出现部分像素信息丢失,影响二维特征的三维转换精度,重建效果与预期效果差距较大,且产生冗余特征信息,严重影响重建效率.故此,本文提出一种基于多尺度特征融合的适合建筑室内三维场景重建的全新的方法.具体实现如下:首先,对建筑室内三维场景的多视点采样图像进行融合;然后,建立采集图像的多尺度特征融合网络;最后,完成基于多尺度特征融合的建筑室内三维场景快速重建.对比实验数据得出:本文提出的多尺度特征融合在三维重建过程中起到了关键作用,三维特征匹配精准度与重建效率得到了显著提升,有效弥补了传统重建方法的不足.
Abstract
The traditional three-dimensional(3D)scene reconstruction methods mostly use laser scanning technology to obtain structure data.As the acquisition equipment has a certain requirement for the environment space,some pixel information is missing during the indoor image acquisition process,affecting the 3D transformation accuracy of two-dimensional(2D)features.As a result,the reconstruction effect and expected effect have a larger gap,and redundant characteristic information is produced,seriously affecting the reconstruction efficiency.Therefore,a new method based on multi-scale feature fusion,suitable for indoor 3D scene reconstruction,was proposed.The specific implementation was as follows:First,the multi-view sampling images of indoor 3D scenes were integrated;then,a multi-scale feature fusion network for image acquisition was established;finally,the fast reconstruction of indoor 3D scenes based on multi-scale feature fusion was completed.The comparison of experimental data shows that the proposed multi-scale feature fusion plays a key role in the 3D reconstruction process,and the accuracy of 3D feature matching and reconstruction efficiency are significantly improved,which effectively solves the shortcomings of traditional reconstruction methods.
关键词
多尺度特征融合/建筑室内/三维场景/快速重建Key words
multi-scale feature fusion/building interior/three-dimensional(3D)scene/fast reconstruction引用本文复制引用
基金项目
重庆市自然科学基金(CSTB2022NSCQ-MSX1615)
出版年
2024
国家科技期刊平台
NSTL
万方数据