Efficient Camera Tampering Detection with Automatic Parameter Calibration
Alexey Sidnev, Marina Konstantinovna Barinova, Sergei Nosov · 2018
Camera anomaly detection, an important part of modern digital security and surveillance cameras, captures malicious actions against the camera like spray paint, occlusion, and displacement. Multiple challenges exist for solving camera anomaly detection in a practical way, such as limited computational budget and difficult tampering events. In this article, we present a fast and accurate approach for camera anomaly detection, which consists of the application of low-level computer vision techniques combined with a unique automatic parameter calibration method that eliminates the need for manual parameter tuning. Based on the results of our experiments, the proposed algorithm detects more than 96 percent of malicious events without false alarms for 13 hours of video with 350 tampering events from more than 20 different scenes. On the TCAD dataset [15], we achieved state-of-the-art results.