摘要
基于标准的2D卷积核的RGBD语义分割模型多是将深度图作为一个单独的通道,由于其卷积核特性的限制,无法充分挖掘深度信息带来的几何结构信息。针对该缺陷,构建深度敏感卷积核和池化层实现对深度信息的丰富挖掘;并使用深度敏感空间金字塔池化对多尺度信息进行提取,实现对不同尺度物体分割的效果。NYU v2和SUN RGB-D数据集上的实验结果表明此方法有效提高了整体的语义分割精度。
The RGBD semantic segmentation model based on the standard 2D convolution kernel mostly takes the depth map as a single channel.Due to the limitation of convolution kernel characteristicsthe geometric structure information brought by the depth information cannot be fully exploited.To overcome this defectthis paper constructs depth-sensitive convolution kernels and a pooling layer to make rich mining of depth information,and uses depth-sensitive spatial pyramid pooling to extract multi-scale informationso as to realize the segmentation of objects of different scales.Results of experiment on NYU v2 and SUN RGB-D datasets show that this method effectively improves the overall semantic segmentation accuracy.
作者
杨胜杰
仇振安
高小宁
李建勋
YANG Shengjie;QIU Zhen'an;GAO Xiaoning;LI Jianxun(Luoyang Institute of Electro-Optical Equipment,AVIC,Luoyang 471000,China;Aviation Military Representative Officeof Army Armament Department in Luoyang,Luoyang 471000,China;Shanghai Jiao Tong University,Shanghai 200240,China)
出处
《电光与控制》
CSCD
北大核心
2020年第12期84-89,共6页
Electronics Optics & Control
基金
国家自然科学基金(61673265)
民用飞机特别研究项目(MJ-2017-S-38)
航空科学基金(20170157001)。
关键词
RGBD语义分割
深度敏感卷积
空间金字塔池化
RGBD semantic segmentation
depth-sensitive convolution
spatial pyramid pooling