Abstract
Anomaly detection in high-density crowds is considered as an important research problem. Detecting anomalous crowd behavior is a complex problem due to unpredictable human behaviors and complex interactions of individuals in groups. In this paper, we present a supervised approach to detect anomalous trajectories. The proposed method has four subsequent steps. In the first step, we extract trajectories from the input video sequence. In the second step, we compute novel features from these trajectories. In the third step, we classify each trajectory into two classes, i.e., anomalous and normal. In the fourth step, we employ a clustering algorithm to cluster all anomalous trajectories. The resultant cluster indicates the anomalous regions in the scene. We evaluated the proposed approach on two publicly available benchmark datasets. From experimental results, we demonstrate the proposed method outperforms other state-of-the-art methods. (C) 2020 INT TRANS J ENG MANAG SCI TECH.