School of Computer Science and Technology, Shandong University of Technology, Zibo 255000, China
Show less
文章历史+
Received
Published
Revised
2025-03-04
2026-02-28
2025-11-21
Online Date
2026-05-30
PDF (10636K)
摘要
为解决室内暗光场景下视觉定位存在的定位耗费时间多、拍摄图像质量差,使图像特征匹配效果差,导致定位结果不准确等问题,提出了一种基于暗光图像增强的SFM室内视觉定位算法。在离线阶段,首先通过视频进行图像数据采集;其次利用即从运动恢复结构(structure from motion,SFM)中的分解本质矩阵就可以得到两张图像的相对位置,重构初始点云;再通过PnP算法求解出所有图像关于初始零点的相对位置,获得离线图像的位置信息和比例尺;最终得到由图像、位置信息以及比例尺信息构成的离线数据库。在线测试阶段,首先采集测试图像;其次使用了一种基于KANs的改进去噪网络扩散模型暗光图像增强算法(KANDiff)对拍摄的暗光图像进行恢复,并对恢复后的图像进行PnP定位,再利用比例尺得到用户实时定位信息。实验结果表明,该系统获得了准确的定位结果,平均定位误差为0.17 m。该算法提高了暗光条件下的定位精度。
Abstract
To address the challenges of prolonged localization time, poor image quality, and low feature matching accuracy in low-light indoor visual localization, an indoor visual localization algorithm based on SFM (structure-from-motion) and low-light image enhancement was proposed. In the offline phase, image data were collected via video recording. The essential matrix was decomposed in the SFM framework to obtain the relative poses between image pairs, enabling the reconstruction of an initial point cloud. The PnP (perspective-n-point) algorithm was then applied to estimate the relative poses of all images with respect to an initial reference point, yielding their position and scale information. An offline database consisting of images, positions, and scale information was constructed. During the online phase, test images were captured and enhanced using a denoising diffusion model based on kernel KANs (adaptive networks), referred to as KANDiff. The enhanced images were then used for localization via the PnP algorithm, and the real-time user position was estimated using the computed scale. Experimental results demonstrated that the proposed method achieved an average localization error of 0.17 meters. It is concluded that localization accuracy under low-light conditions can be significantly improved using the proposed approach.
由于对精准位置服务的需求日益增长,室内定位技术得到了快速发展,研究者围绕定位精度、鲁棒性和部署成本等方面提出了多种改进方案。文献[5]面向到达时间差(time difference of arrival,TDOA)场景,将智能优化算法嵌入到室内TDOA定位模型中,通过优化超参数与测距误差分布,提升了基于时间差测距的定位精度与稳定性。为改善传统WIFI指纹定位在高维特征建模与实时推理中的精度与复杂度矛盾,文献[6]构建了基于二维卷积神经网络的WIFI室内定位模型,通过卷积特征自动提取和端到端回归,提高了指纹匹配的鲁棒性。
在暗光环境下捕获的图像,其质量往往因信噪比降低、细节损失及色彩失真等因素而显著下降。为解决暗光场景下图像的质量退化问题,使用了一种基于KANs的改进去噪网络扩散模型暗光图像增强算法(KANDiff)。此方法在Özdenizci等[24]的基础上,将两个网络进行有效结合,如图2所示。整体结构遵循扩散模型前向加噪后向去噪的形式,现有的扩散模型大多针对固定尺寸的图像进行图像恢复。然而,本文模型通过在前向加噪阶段将图像划分为N个样本补丁,进行扩散,并在采样阶段将这些补丁合并,从而实现对多种尺寸图像的恢复。此方法在网络中引入跳跃连接,以确保所提取的特征包含更丰富的语义信息。此外,由于传统噪声估计网络无法满足图像恢复中对噪声预测的需求,将 KANs 引入噪声估计网络,以发挥 KAN 的生成能力,提升网络的高效性。
WangQianqian, XiongYuan, JiangHan, et al. Visual localization of UAV images with scene appearance change[J]. Aerospace Control and Application, 2024, 50(1): 56-67.
[4]
ZhangT, GuoS, MaL, et al. Hierarchical image retrieval method based on bag-of-visual-word and eight-point algorithm with feature clouds for visual indoor positioning[C]// 27th Asia Pacific Confe-rence on Communications (APCC). New York: IEEE, 2022: 455-459.
[5]
KangZ, YangJ, YangZ, et al. A review of techniques for 3D reconstruction of indoor environments[J]. ISPRS International Journal of Geo-Information, 2020, 9(5): 1-31.
LiShuaichen, WuJianfeng. Research on indoor positioning based on intelligent optimization algorithm and optimized BP neural network[J]. Science Technology and Engineering, 2024, 24(20): 8568-8576.
ZhangTianying, ShiMingquan, CuiLizhen, et al. 2D convolutional neural network based WIFI indoor localization system[J]. Science Technology and Engineering, 2023, 23(28): 12168-12174.
[10]
ChenC, WangB, LuC X, et al. Deep learning for visual localization and mapping: a survey[J]. IEEE Transactions on Neural Networks and Learning Systems, 2024, 35(12): 17000-17020.
[11]
JiaS, MaL, YangS, et al. A novel visual indoor positioning method with efficient image deblurring[J]. IEEE Transactions on Mobile Computing, 2022, 22(7): 3757-3773.
HeFaxu, LüMengyan, ZhangKunpeng, et al. Indoor visual localization system based on blurred images[J]. Science Technology and Engineering, 2024, 24(32): 13911-13924.
[14]
HyeonJ, JangB, ChoiH, et al. Photo-realistic 3D model based accurate visual positioning system for large-scale indoor spaces[J]. Engineering Applications of Artificial Intelligence, 2023, 123. DOI: 10.1016/j.engappai.2023.106256.
[15]
GaoH, DaiK, WangK, et al. ALNet: an adaptive channel attention network with local discrepancy perception for accurate indoor visual localization[J]. Expert Systems with Applications, 2024, 250.DOI: 10.1016/j.eswa.2024.123792.
[16]
JiangY, GongX, LiuD, et al. Enlightengan: deep light enhancement without paired supervision[J]. IEEE Transactions on Image Processing, 2021, 30: 2340-2349.
[17]
WeiC, WangW, YangW, et al. Deep retinex decomposition for low-light enhancement[J]. Tutorial on Gabor Filters, 2002, 40: 1-23.
[18]
RenX, YangW, ChengW H, et al. LR3M: robust low-light enhancement via low-rank regularized retinex model[J]. IEEE Transactions on Image Processing, 2020, 29: 5862-5876.
[19]
HaoS, HanX, GuoY, et al. Low-light image enhancement with semi-decoupled decomposition[J]. IEEE Transactions on Multimedia, 2020, 22(12): 3025-3038.
[20]
LüF, LiY, LuF. Attention guided low-light image enhancement with a large scale low-light simulation dataset[J]. International Journal of Computer Vision, 2021, 129(7): 2175-2193.
[21]
MaL, MaT, LiuR, et al. Toward fast, flexible, and robust low-light imageenhancement[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2022: 5637-5646.
WangDengfei, SuHongsheng, ChenGuangwu, et al. Real time low light enhancement method of TEDS images based on focused linear attention Retinexformer[J]. Journal of Railway Science and Engineering, 2024, 21(11): 4840-4850.
[24]
SunJ, XieY, ChenL, et al. Neuralrecon: real-time coherent 3D reconstruction from monocular video[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2021: 15598-15607.
[25]
SunJ, ChenX, WangQ, et al. Neural 3D reconstruction in the wild[C]// ACM SIGGRAPH 2022 Conference Proceedings. New York: ACM, 2022: 1-9.
[26]
ZhouZ, TulsianiS. Sparsefusion: distilling view-conditioned diffusion for 3d reconstruction[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2023: 12588-12597.
[27]
AnciukeviciusT, XuZ, FisherM, et al. Renderdiffusion: image diffusion for 3D reconstruction, inpainting and generation[C]// Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. New York: IEEE, 2023: 12608-12618.
QiXian, LiuDaming, ChangJiaxin. Multi-view 3D reconstruction based on improved self-attention mechanism[J]. Computer and Modernization, 2024(11): 106-112.
[30]
ÖzdenizciO, LegensteinR. Restoring vision in adverse weather conditions with patch-based denoising diffusion models[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(8): 10346-10357.
[31]
BertonG, TrivignoG, CaputoB, et al. Eigenplaces: training viewpoint robust models for visual place recognition[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. New York: IEEE,2023: 11080-11090.
[32]
XueF, WuX, CaiS, et al. Learning multi-view camera relocali-zation with graph neural networks[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2020: 11375-11384.
[33]
WangB, ChenC, LuC X, et al. Atloc: attention guided camera localization[C]// Proceedings of the AAAI Conference on Artificial Intelligence. New York: AAAI, 2020, 34(6): 10393-10401.
[34]
ZhouQ, SattlerT, PollefeysM, et al. To learn or not to learn: visual localization from essential matrices[C]// IEEE International Conference on Robotics and Automation (ICRA). New York: IEEE, 2020: 3319-3326.
[35]
ChenS, LiX H, WangZ R, et al. Dfnet: enhance absolute pose regression with direct feature matching[C]// Proceedings of European Conference on Computer Vision. Cham: Springer, 2022: 1-17.