敵対的生成ネットワークを用いた動画像異常検知における背景除去の有用性の検討

抄録

Research on anomaly detection has been conducted to capture anomalies in various situations in daily life. In this study, we aim to build a more accurate anomaly detection system that can detect people and cars behaving abnormally in the video. The background information is considered to be unnecessary for anomaly detection. Therefore, we propose a method to multiply the difference between the generated image and the true value of the optical flow by the image with background removal. We confirm that detection results with high accuracy are produced compared with the conventional method.