Abnormal event detection with semi-supervised sparse topic model

Most research on anomaly detection has focused on event that is different from its spatial-temporal neighboring events. However, it is still a significant challenge to detect anomalies that involve multiple normal events interacting in an unusual pattern. To address this problem, a novel semi-supervised method based on sparse topic model is proposed to detect anomalies in video surveillance. Short local trajectory method is used to extract motion information in order to improve the robustness of trajectories. For the purpose of strengthening the relationship of interest points on the same trajectory, the Fisher kernel method is applied to obtain the representation of trajectory which is quantized into visual word. Then, the sparse topic model is proposed to explore the latent motion patterns and achieve a sparse representation for the video scene. Finally, a semi-supervised learning method is applied to enhance the discrimination of model and improve the performance of anomaly detection. Experiments are conducted on QMUL dataset and AVSS dataset. The results demonstrated the superior efficiency of the proposed method.