Deep Learning-Based Data Drift Detection of Image Rotation in Surveillance Videos

Surveillance systems are actively being in use for public safety. Using CCTV video data, such tasks as object tracking, and face recognition can be performed. However, if surveillance equipment rotates due to collisions with objects or natural events like typhoons, rotated image fed into the network can degrade the performance of the network model, making it necessary to detect data drift in a timely manner. In this paper, we propose a neural network-based model for detecting data drift due to image rotation. When two images are given to the network as input, the model detects whether there is image rotation drift or not. The experiment result of the model using the UCSD Ped1 and Ped2 datasets achieved significantly high accuracy.