Video Surveillance for Indoor Office Environment Based on Object-Level Anomaly Detection

Abstract Traditional methods of Abnormal Behavior Detection (ABD) process the surveillance video on frame-level, which ignores object-level abnormal behavior patterns. To address problem, this paper presents Object-Level Anomaly model (OLAD), aims to various normal patterns different objects and interaction between them. Specifically, OLAD introduces an encoding-embedding network transform information into feature space. By integrating such information, processes both frame-level cues in for ABD. In addition, we construct our own dataset Northking2022 especially office scenes because lack public datasets indoor environments. Experimental results show that gains better performance benchmark Northking2022.