Privacy-Preserving Smart Surveillance with Cross-Dataset Violence Detection and Decentralized Evidence Governance

2026-05-31

AI Summary

This study addresses critical limitations in conventional intelligent surveillance systems, which suffer from privacy vulnerabilities due to centralized video storage and exhibit poor generalization of violence detection models across datasets. To overcome these issues, the authors propose a privacy-preserving surveillance framework that decouples violence detection from evidence access. A lightweight MobileNetV2-BiLSTM model achieves high-accuracy temporal violence recognition (93.5% accuracy and 0.980 ROC-AUC on combined test sets) and triggers immediate encryption of suspicious segments. Decryption is governed by a decentralized evidence management mechanism combining Shamir’s secret sharing, public-key cryptography, time-limited tokens, and two-factor authentication, requiring multi-threshold authorization. This work pioneers the integration of cross-dataset robust detection with auditable, privacy-aware access control and identifies significant distributional shift in the RWF-2000 dataset.

Abstract

AI-enabled surveillance can accelerate public-safety response, yet most systems still leave recorded evidence under centralized administrative control. This paper proposes a privacy-preserving smart surveillance framework that separates incident detection from evidence disclosure. A lightweight MobileNetV2-based video classifier detects violent clips, while each recorded incident segment is immediately encrypted and made accessible only through threshold-based approval. The decryption key is split with Shamir's Secret Sharing, member shares are protected with public-key cryptography, and voting is supported by time-limited tokens, two-factor authentication, signatures, and audit logs. This study evaluates MobileNetV2+LSTM, MobileNetV2+BiLSTM, and MobileNetV2+temporal CNN heads on SCVD, RWF-2000, and Real-Life Violence Situations under seven in-domain and cross-dataset scenarios. The best all-source model, MobileNetV2+BiLSTM, reaches 93.5% test accuracy and ROC-AUC 0.980% on the merged held-out set, while lower RWF-2000 slice performance confirms persistent dataset shift.

Problem

• privacy-preserving surveillance
• decentralized evidence governance
• cross-dataset violence detection
• smart surveillance
• evidence disclosure

Innovation

• privacy-preserving surveillance
• cross-dataset violence detection
• decentralized evidence governance
• Shamir's Secret Sharing
• MobileNetV2+BiLSTM