Secrets in Motion: Privacy-Preserving Video Classification with Built-In Access Control

This work addresses the challenges in secure video classification and access by proposing SIM, a fine-grained access control-based multi-party framework. Leveraging Ciphertext Policy Attribute-Based Encryption (CP-ABE) and Multi-Key Homomorphic Encryption (MKHE), SIM ensures privacy and security for selectively encrypted videos, even in scenarios involving untrusted cloud providers. Our construction considers an intelligent surveillance system where multiple cameras independently encrypt data about identified objects in a video stream, enabling arbitrary cloud computations without direct camera-technician interaction. Additionally, unlike standard MKHE schemes, SIM eliminates the decryption phase, enhancing security. Finally, for cloud classification, SIM integrates a Machine Learning (ML) model with MKHE evaluation, enabling multiple parties to collaborate securely without trusting each other completely. The use of MKHE and CP-ABE enables SIM to support object of interest search, which facilitates efficient and secure search over classified encrypted videos stored with an untrusted cloud.