Anomaly Detection in Videos Recorded by Drones in a Surveillance Context
Jordan Henrio, Tomoharu Nakashima · 2018
Automatic anomaly detection is an important difficulty in various domains such as surveillance, medicine and industry. Current surveillance systems generally use fixed-position cameras that cannot track anomalies. By virtue of recent industrial developments, drones are increasingly employed in various domains. They can also be employed for surveillance as moving cameras. Nevertheless, the recent literature lacks contributions describing anomaly detection in videos recorded by drones or for those with dynamic backgrounds. This paper presents specific examination of anomaly detection on the mini-drone video dataset which consists of surveillance videos recorded by a drone. The suggested anomaly detector is a deep neural network composed of a convolutional neural network and a recurrent neural network, trained using supervised learning. Experiments were conducted on both the mini-drone video dataset and the more popular UMN dataset. Although our model achieves state-of-the-art results on the UMN dataset, it is less data-efficient. However, the main goal of this study is to promote research efforts to resolve the difficulty of anomaly detection in videos with dynamic background and to provide first baseline results for additional contributions on the mini-drone video dataset.