Modeling the Trade-off of Privacy Preservation and Activity Recognition on Low-Resolution Images
2023年1月1日·
,
,
,
,
,
,
,
,
,
·
0 分钟阅读时长
Yuntao Wang
Zirui Cheng
Xin Yi
Yan Kong
Xueyang Wang
Xuhai Xu
Yukang Yan
Chun Yu
Shwetak Patel
Yuanchun Shi
摘要
采用低分辨率图像传感器的计算机视觉系统，既可提供活动识别等智能服务，又能在硬件层面减少不必要的视觉隐私泄露。然而，隐私保护与高精度机器识别对图像分辨率存在天然的对抗性需求：分辨率越低，隐私风险越小，但识别性能也可能受损。因此，对隐私保护与机器识别性能之间的权衡进行建模，有助于指导未来面向低分辨率传感器的隐私保护视觉系统设计。本文以居家日常生活活动（ADLs）为场景，首先通过用户调研识别并提炼关键的视觉隐私特征；随后系统量化并分析图像分辨率变化对人类与机器在活动识别与隐私感知任务中的表现影响；进一步考察现代图像超分辨率技术在多大程度上会改变上述关系。基于实验结果，我们提出了一种用于在低分辨率图像上建模隐私保护与活动识别性能权衡的方法。
类型
出版物
Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems
Authors
Associate Professor (Research Track)
Yuntao Wang’s research centers on physiobehavioral computing and
intelligent interaction for mobile and wearable systems. His work
focuses on (1) developing robust, efficient sensing that performs
reliably on mainstream devices, (2) extracting spatiotemporal patterns
from multimodal signals to infer interaction intent by leveraging
natural behavioral correlations, and (3) designing edge-efficient interfaces
that deliver high performance on mobile and wearable platforms.
He has published 90+ papers, received 10 international conference awards,
and holds 30+ granted patents. His contributions have been
recognized with honors including the Wu Wenjun AI Outstanding Youth
Award (2024), the CAST Young Elite Scientists Sponsorship Program (2022),
the Qinghai High-Level Innovation & Entrepreneurship Leading Talent (2024),
and the First Prize of the China Electronics Institute Science & Technology Award (2019).