摘要
进入大数据时代后,互联网应用和信息服务全面普及,大量的个人敏感生物信息被收集整理,导致隐私泄露风险增加;事件相机作为新型的生物启发式传感器,具有低延迟、高动态、无纹理等特性,可为解决数据端隐私保护问题提供全新的技术途径,也因其光敏工作原理而适用于家庭监控等私人场景。本文从大数据时代的隐私泄露问题、事件相机在隐私保护中的优势两方面,深入分析了事件相机用于隐私保护的研究背景;系统梳理了基于人脸模板的隐私保护、基于去识别的隐私保护、基于点云混沌加密的隐私保护等传统的敏感生物信息隐私保护方法,包括行人重识别、手势识别、面部分析在内的面向隐私保护的事件感知方法以及两大类方法的研究进展。进一步总结了强度图像重建、图像修复、视频重建等基于事件流的图像重建与修复新进展,完成了基于6种算法的图像重建及其结果分析,证明已有重建算法对纹理信息的恢复能力有限,反向验证了基于事件相机的隐私保护技术可行性。针对事件相机未来的规模化应用,提出了降低硬件成本、改进算法网络、从市场角度推动等发展建议,以期为事件相机的隐私保护深化应用提供基础参考。
In the era of big data,the widespread adoption of Internet applications and information services has resulted in the extensive collection of individuals'sensitive biological information,increasing the risk of privacy breaches.Event cameras,as novel bio-inspired sensors,exhibit characteristics such as low latency,high dynamics,and texture independence.They offer a fresh technological approach to addressing privacy protection issues on the data side,making them suitable for private applications like home monitoring.This study thoroughly analyzes the research background of using event cameras for privacy protection,focusing on the privacy leakage issues of the big data era and the advantages of event cameras in privacy protection.It systematically reviews traditional methods for protecting sensitive bioinformation privacy,including face-template-based privacy protection,deidentification-based privacy protection,and privacy protection based on point cloud chaotic encryption.Additionally,it has examined the research progress in privacy-preserving event perception methods,including pedestrian re-identification,gesture recognition,and facial analysis.Further,the study also summarizes advancements in event-based image reconstruction and restoration,including intensity image reconstruction,image restoration,and video reconstruction,based on six algorithms.The results demonstrated that existing reconstruction algorithms have limited capability in recovering texture information,and the feasibility of the privacy protection technology based on event cameras is confirmed.For the future scaled-up application of event cameras,development recommendations are proposed,including reducing hardware costs,improving algorithm networks,and driving initiatives from a market perspective,aiming to provide a foundational reference for the deepened application of privacy protection using event cameras.
作者
沈艳晴
董鹏飞
张恙菁
陈仕韬
郑南宁
Shen Yanqing;Dong Pengfei;Zhang Yangjing;Chen Shitao;Zheng Nanning(National Key Laboratory of Human-Machine Hybrid Augmented Intelligence,Xi’an 710049,China;National Engineering Research Center for Visual Information and Applications,Xi’an 710049,China;Institute of Artificial Intelligence and Robotics,Xi’an Jiaotong University,Xi’an 710049,China)
出处
《中国工程科学》
CSCD
北大核心
2024年第1期202-215,共14页
Strategic Study of CAE
基金
中国工程院咨询项目“混合智能及产业集群发展战略研究”(2022-PP-07)。
关键词
智能系统
事件相机
隐私保护
敏感生物信息
intelligent system
event camera
privacy protection
sensitive bioinformation
作者简介
通讯作者:郑南宁,西安交通大学人工智能与机器人研究所教授,中国工程院院士,研究方向为无人驾驶、计算机视觉、具身智能,E-mail:nnzheng@mail.xjtu.edu.cn。
