Video anomaly detection using knowledge distillation

With advent of technology, tremendous size of visual data is being generated by video surveillance systems, which makes harder to search, analyze, and detect anomalies on video data by human operators. Consequently, video anomaly detection has become a vital task for smart video surveillance systems because of it's significant potential to minimize the video data to be analysed by choosing unusual and critical patterns in the scenes. Detecting anomaly patterns in videos is a challenging task due to complex scenes, huge diversity of anomalies, and fuzzy nature of the task. In this thesis, we introduce three novel relational knowledge distillation methods and three novel ensemble based knowledge distillation methods to handle both robust detection of anomalous events and gradual adaptation to different anomaly patterns in new videos while not forgetting anomaly patterns learned from the previous video data. Our relational knowledge distillation methods utilize feature and relation based knowledge distillation method with a unique attention mechanism to provide rich and structured information transfer from teacher to student model. Additionally, our ensemble based knowledge distillation methods leverages the adaptation process by providing information transfer from multiple teacher models with different network structures. The proposed knowledge distillation methods are implemented on two state-of-the-art anomaly detection models. We extensively evaluate our methods on two public video anomaly datasets and present detailed analysis of our results.