Object Encryption and Decryption for Human Privacy

The widespread deployment of surveillance cameras and digital media platforms has intensified concerns about unauthorized capture, storage, and misuse of personally identifiable visual information. This paper presents a real-time privacy-preservation system that integrates YOLOv8l-based person detection with ChaCha20 stream cipher encryption and median blur visual anonymization for video data. The proposed pipeline processes video frames using OpenCV, detects persons using the YOLOv8l model with a confidence threshold of 0.4 and IoU of 0.5, encrypts each unique person’s frame using a 256-bit ChaCha20 key with a 128-bit nonce, and simultaneously blurs the detected person region using a 25×25 median blur kernel. A centroid-based grid rounding mechanism (80px grid) ensures stable person tracking across frames, while a time-interval constraint limits re-encryption to once per second per person, significantly reducing redundant encrypted frame generation from hundreds of frames down to 1–6 frames per unique person per video. Each encrypted frame is assigned a unique six-character alphanumeric identifier for selective authorized decryption. A side-by-side visualization module enables direct comparison of the anonymized public view against the decrypted administrative view. The system achieves a person detection accuracy of 88.3%, decryption latency of 0.0012 seconds per frame, and operates in real-time using GPU-accelerated or CPU-based inference without requiring cloud connectivity or specialized hardware. Experimental results confirm that the proposed layered approach provides a practical, storage-efficient, and ethically compliant solution for privacy-preserving visual media management.