袁飞1 饶娆1 周光海2 陈伟斌1
（1.广东技术师范大学自动化学院，广东 广州 510450
2.广州南方测绘科技股份有限公司，广东 广州 510663）
摘要：针对远距离室内指示牌检测准确率较低的问题，提出一种结合超分重建的室内指示牌检测方法。分析单目相机透视成像模型发现，指示牌成像像素随拍摄距离增大而减小；通过对含有室内指示牌的图像进行超分重建，提高指示牌图像分辨率，进而提升指示牌的检测准确率。在SVHN数据集和实验室数据集下，采用SRCNN、SRResNet和SRGAN超分重建神经网络模型进行2倍和4倍超分重建的图像与未超分重建的图像进行对比，指示牌检测准确率均有明显提高。
关键词：超分重建；室内指示牌检测；相机透视成像模型；指示牌成像计算方法；目标检测算法
中图分类号：TP391.41 文献标志码：A 文章编号：1674-2605(2025)04-0005-08
DOI：10.12475/aie.20250405 开放获取
Indoor Signage Detection Method with Super-resolution Reconstruction
YUAN Fei1 RAO Rao1 ZHOU Guanghai2 CHEN Weibin1
(1.School of Automation, Guangdong Polytechnic Normal University, Guangzhou 510450, China
2.Guangzhou Southern Surveying and Mapping Technology Co., Ltd., Guangzhou 510663, China)
Abstract: To address the low detection accuracy of distant indoor signage, this paper proposes an indoor signage detection method with super-resolution reconstruction. Analysis of the monocular camera perspective imaging model reveals that the imaging pixels of signage decrease with increasing shooting distance. By applying super-resolution reconstruction to images containing indoor signage, the resolution of signage images is enhanced, thereby improving detection accuracy. Comparative experiments on the SVHN dataset and a laboratory dataset demonstrate that images reconstructed at 2× and 4× super-resolution using SRCNN, SRResNet, and SRGAN super-resolution reconstruction neural network model significantly outperform non-reconstructed images in signage detection accuracy.
Keywords: super-resolution reconstruction; indoor signage detection; camera perspective imaging model; signage imaging calculation method; object detection algorithms
