Real-Time Adaptive Camera Tamper Detection for Video Surveillance

Low-complexity camera tamper detection based on edge information

“…For all kinds of tamper attacks, the proposed algorithm shows 100% of detection rate with false alarm rate of 2.08%. Tamper detection algorithm based on DFT and HPF in [1] showed DR of 95%, 91.7%, and 82.9% for Covered, Moved, and Defocused camera events, respectively. Although this algorithm has zero FAR, the complexity for detection is quite large and the detection rate of Defocused camera event is not acceptable.…”

“…Defocused camera event is the case which the focal length of the camera is changed to blur the captured video sequences resulting in the reduction of edge pixels. Saglam detected the camera tamper attacks by using the difference of high frequency components after Discrete Fourier Transform (DFT) and High Pass Filtering (HPF) [1]. Aksay detected Covered and Defocused camera events by using wavelet transform and reduced false alarm rate by using edge data [2].…”

Unified Camera Tamper Detection Based on Edge and Object Information

“…The variety of camera tamper attacks can be classified into three types: defocused, covered and moved camera events [ 8 – 10 ]. A defocused camera event is one in which the focal length of the camera is changed to cause blurring of the captured video; a covered camera event is one in which the camera lens is partially or totally occluded by external objects; a moved camera event refers to a situation in which the camera viewing angle is abruptly changed by external forces or the camera is titled by wind or an earthquake.…”

“…Most previous studies on tamper detection have compared features of the current frame with those of the background frames [ 8 , 10 – 12 ]. Features extracted from an image frame include the intensity histogram, edge map and high frequency components after the wavelet transform (WT), the discrete Fourier transform (DFT) or the discrete cosine transform (DCT).…”

“…Although a unified tamper detection criterion was proposed in our previous study [ 15 ], based on the edge changed rate, and in [ 9 ], based on global and local edge energies, the methods used in these studies were still susceptible to scene dynamics. Furthermore, most studies have used quite many thresholds of features for tamper detection, which makes it difficult for the algorithms to be applied to different environmental conditions with optimally-configured thresholds [ 8 , 9 , 11 – 13 , 16 ]. In terms of performance, although most studies have tried to reduce the false alarm rate, experiments were performed only for very short video sequences, which cannot reflect the real video surveillance environment, which has such characteristics as illumination change over 24-h period, scene dynamics, etc .…”