A Novel Hybrid Approach to Detect Camera Tampering

Camera Tampering gained immense interest among researchers with an increased number of cameras deployed in both public and private places by different agencies. Camera Tampering Detection can be used widely to help security officers in their scrutiny of the deployed camera in real time. In this research study we propose a 2-stage approach, in the first stage, rule-based method have been used to detect movement of the camera and in the second stage neural network-based model have been used to detect defocussing or covering of the camera. Foreground have been extracted from the background subtraction method, which is then used to determine the percentage of area covered by foreground objects. Finally, a threshold has been used to determine the movement of the camera. In neural network-based method, Convolutional Neural Network based MobilenetV2 architecture have been used to separate the blur and defocussed tampering from normal camera view. Our research study shows that combined methods provide state-of-the art performance with around 97% of F1 score and properly detect different types of camera tampering.