Invisibility Cloak: Personalized Smartwatch-Guided Camera Obfuscation
summary
System links smartwatch IMU signals to on-camera motion to remove a person’s pixels at the edge before streaming to cloud services, using a contrastive embedding that aligns 2‑second IMU windows with short video motion tracks and then assigns devices to people for per-person obfuscation. The pipeline reports 95.48% accuracy among 10 users with 2 seconds of data and relies on a contrastive embedding plus local edge processing to apply configurable obfuscation levels.
AI-written summary · about
chapters (11)
- 0:00Introduction to Invisibility Cloak
- 1:09Privacy challenges in vision AI
- 2:11Limitations of existing approaches
- 2:56Invisibility Cloak system overview
- 3:37Motion-based user association
- 4:42Contrastive learning pipeline
- 7:28Device association with Hungarian algorithm
- 8:10Data collection and synthesis
- 9:34Evaluation and results
- 10:15Limitations and future work
- 10:37Q&A session
related talks
40mNew Face, Who Dis? Protecting Privacy in a World of Surveillance
20mEnhancing Mobile Content Privacy with Proxemics Aware Notifications and Protection
10mGO-Finder: A Registration-Free Wearable System for Assisting Users in Finding Lost Objects via ...
20mWIP: Blurred Lines -- A GDPR-Compliant Framework for Anonymising Automotive...
21mCombining Privacy and AI
11mPrivacy-Preserving Protocols for Smart Camera Systems and Other IoT Devices
32mPrivacy Protection. Privacy by Design: Strengthening Compliance
19mGDPR, AI & use of Video data