Journal of Computer Applications

    Next Articles

Privacy preserving localization of surveillance images based on large vision models

  

  • Received:2024-10-31 Revised:2024-12-03 Accepted:2024-12-04 Online:2025-01-06 Published:2025-01-06

基于视觉大模型隐私保护的监控图像定位

李强1,白少雄1,熊源2,袁薇3*   

  1. 1.国能运输技术研究院有限责任公司,北京 100000; 2.中山大学 网络空间安全学院,深圳 广东 518107; 3.中国软件评测中心(工业和信息化部软件与集成电路促进中心),北京 102206


  • 通讯作者: 袁薇

Abstract: Visual localization of surveillance images is an important technology in industrial intelligence. Existing visual localization algorithms lack the protection of the privacy information in the image and may lead to the leakage of sensitive content in data transmission. To addresses the problem of using large vision models (LVMs) in visual localization,a privacy preserving localization method of surveillance images based on Large Vision Models (LVMs) was proposed. Firstly, the architecture of privacy preserving LVM based visual localization was designed to transfer the style of input images by using a few prompts and reference images. Then, a feature-matching algorithm for image style transfer was designed to estimate the camera pose. Experimental results on public datasets show that the localization error of the proposed algorithm is less than 0.25 m, significantly reducing the privacy leakage meanwhile ensuring localization accuracy.

Key words: diffusion model, surveillance localization, Large Vision Model (LVM), visual localization, privacy preserving

摘要: 监控图像的视觉定位是工业智能领域的关键技术。针对现有视觉定位算法缺少对图像中隐私信息的保护,在数据传输过程中容易导致敏感内容泄露的问题,提出一种基于视觉大模型(LVMs)的监控图像定位方法。首先,设计基于大模型隐私保护的视觉定位架构,利用少量文本提示和参考图像对输入图像进行风格迁移;其次,提出面向风格迁移图像的特征匹配算法用于估计相机位姿。在公开数据集上的实验结果表明,所提算法定位结果误差小于0.25 m,在保证定位精度的前提下大幅减少了隐私泄露。

关键词: 扩散模型, 监控定位, 视觉语言模型, 视觉定位, 隐私保护

CLC Number: