Moving object detection in the encrypted domain

The privacy-preserving moving object detection has drawn a lot of interest lately. Nevertheless, current approaches use Paillier's scheme for encryption that impractical in real-time applications due to high computational complexity. In addition, none of them are fully compatible with popular background modeling methods. In this paper, a fast and secure encryption scheme for a surveillance system has been proposed. The algorithm allows the detection of a moving object to be implemented directly in the encryption domain. The proposed scheme separates every pixel into two parts. The first part of a pixel (most significant bits) is scrambled to encrypt the image, and the second part of the pixel (least significant bits) remains unchanged. This strategy allows the proposed encryption scheme to be compatible with the mixture of Gaussians (GMM) that is one of the most widely used background modeling methods to detect moving objects. The proposed scheme requires low computations and produces almost the same detection result as the GMM when it is applied to unencrypted videos. Security analysis of the proposed method also proves the robustness of the encryption process.